/src/libavif/src/reformat.c
Line | Count | Source |
1 | | // Copyright 2019 Joe Drago. All rights reserved. |
2 | | // SPDX-License-Identifier: BSD-2-Clause |
3 | | |
4 | | #include "avif/internal.h" |
5 | | |
6 | | #include <assert.h> |
7 | | #include <stdint.h> |
8 | | #include <string.h> |
9 | | |
10 | | #if defined(_WIN32) |
11 | | #include <process.h> |
12 | | #include <windows.h> |
13 | | #else |
14 | | #include <pthread.h> |
15 | | #endif |
16 | | |
17 | | static void * avifMemset16(void * dest, int val, size_t count) |
18 | 0 | { |
19 | 0 | uint16_t * dest16 = (uint16_t *)dest; |
20 | 0 | for (size_t i = 0; i < count; i++) |
21 | 0 | *dest16++ = (uint16_t)val; |
22 | 0 | return dest; |
23 | 0 | } |
24 | | |
25 | | struct YUVBlock |
26 | | { |
27 | | float y; |
28 | | float u; |
29 | | float v; |
30 | | }; |
31 | | |
32 | | avifBool avifGetRGBColorSpaceInfo(const avifRGBImage * rgb, avifRGBColorSpaceInfo * info) |
33 | 2.37k | { |
34 | 2.37k | AVIF_CHECK(rgb->depth == 8 || rgb->depth == 10 || rgb->depth == 12 || rgb->depth == 16); |
35 | 2.37k | if (rgb->isFloat) { |
36 | 0 | AVIF_CHECK(rgb->depth == 16); |
37 | 0 | } |
38 | 2.37k | if (rgb->format == AVIF_RGB_FORMAT_RGB_565) { |
39 | 0 | AVIF_CHECK(rgb->depth == 8); |
40 | 0 | } |
41 | | // Cast to silence "comparison of unsigned expression is always true" warning. |
42 | 2.37k | AVIF_CHECK((int)rgb->format >= AVIF_RGB_FORMAT_RGB && rgb->format < AVIF_RGB_FORMAT_COUNT); |
43 | | |
44 | 2.37k | info->channelBytes = (rgb->depth > 8) ? 2 : 1; |
45 | 2.37k | info->pixelBytes = avifRGBImagePixelSize(rgb); |
46 | | |
47 | 2.37k | info->offsetBytesR = 0; |
48 | 2.37k | info->offsetBytesG = 0; |
49 | 2.37k | info->offsetBytesB = 0; |
50 | 2.37k | info->offsetBytesA = 0; |
51 | 2.37k | info->offsetBytesGray = 0; |
52 | | |
53 | 2.37k | switch (rgb->format) { |
54 | 0 | case AVIF_RGB_FORMAT_RGB: |
55 | 0 | info->offsetBytesR = info->channelBytes * 0; |
56 | 0 | info->offsetBytesG = info->channelBytes * 1; |
57 | 0 | info->offsetBytesB = info->channelBytes * 2; |
58 | 0 | break; |
59 | 682 | case AVIF_RGB_FORMAT_RGBA: |
60 | 682 | info->offsetBytesR = info->channelBytes * 0; |
61 | 682 | info->offsetBytesG = info->channelBytes * 1; |
62 | 682 | info->offsetBytesB = info->channelBytes * 2; |
63 | 682 | info->offsetBytesA = info->channelBytes * 3; |
64 | 682 | break; |
65 | 0 | case AVIF_RGB_FORMAT_ARGB: |
66 | 0 | info->offsetBytesA = info->channelBytes * 0; |
67 | 0 | info->offsetBytesR = info->channelBytes * 1; |
68 | 0 | info->offsetBytesG = info->channelBytes * 2; |
69 | 0 | info->offsetBytesB = info->channelBytes * 3; |
70 | 0 | break; |
71 | 0 | case AVIF_RGB_FORMAT_BGR: |
72 | 0 | info->offsetBytesB = info->channelBytes * 0; |
73 | 0 | info->offsetBytesG = info->channelBytes * 1; |
74 | 0 | info->offsetBytesR = info->channelBytes * 2; |
75 | 0 | break; |
76 | 1.68k | case AVIF_RGB_FORMAT_BGRA: |
77 | 1.68k | info->offsetBytesB = info->channelBytes * 0; |
78 | 1.68k | info->offsetBytesG = info->channelBytes * 1; |
79 | 1.68k | info->offsetBytesR = info->channelBytes * 2; |
80 | 1.68k | info->offsetBytesA = info->channelBytes * 3; |
81 | 1.68k | break; |
82 | 0 | case AVIF_RGB_FORMAT_ABGR: |
83 | 0 | info->offsetBytesA = info->channelBytes * 0; |
84 | 0 | info->offsetBytesB = info->channelBytes * 1; |
85 | 0 | info->offsetBytesG = info->channelBytes * 2; |
86 | 0 | info->offsetBytesR = info->channelBytes * 3; |
87 | 0 | break; |
88 | 0 | case AVIF_RGB_FORMAT_RGB_565: |
89 | | // Since RGB_565 consists of two bytes per RGB pixel, we simply use |
90 | | // the pointer to the red channel to populate the entire pixel value |
91 | | // as a uint16_t. As a result only offsetBytesR is used and the |
92 | | // other offsets are unused. |
93 | 0 | info->offsetBytesR = 0; |
94 | 0 | info->offsetBytesG = 0; |
95 | 0 | info->offsetBytesB = 0; |
96 | 0 | break; |
97 | 0 | case AVIF_RGB_FORMAT_GRAY: |
98 | 0 | info->offsetBytesGray = info->channelBytes * 0; |
99 | 0 | break; |
100 | 0 | case AVIF_RGB_FORMAT_GRAYA: |
101 | 0 | info->offsetBytesGray = info->channelBytes * 0; |
102 | 0 | info->offsetBytesA = info->channelBytes * 1; |
103 | 0 | break; |
104 | 0 | case AVIF_RGB_FORMAT_AGRAY: |
105 | 0 | info->offsetBytesA = info->channelBytes * 0; |
106 | 0 | info->offsetBytesGray = info->channelBytes * 1; |
107 | 0 | break; |
108 | | |
109 | 0 | case AVIF_RGB_FORMAT_COUNT: |
110 | 0 | return AVIF_FALSE; |
111 | 2.37k | } |
112 | | |
113 | 2.37k | info->maxChannel = (1 << rgb->depth) - 1; |
114 | 2.37k | info->maxChannelF = (float)info->maxChannel; |
115 | | |
116 | 2.37k | return AVIF_TRUE; |
117 | 2.37k | } |
118 | | |
119 | | avifBool avifGetYUVColorSpaceInfo(const avifImage * image, avifYUVColorSpaceInfo * info) |
120 | 2.37k | { |
121 | 2.37k | AVIF_CHECK(image->depth == 8 || image->depth == 10 || image->depth == 12 || image->depth == 16); |
122 | 2.37k | AVIF_CHECK(image->yuvFormat >= AVIF_PIXEL_FORMAT_YUV444 && image->yuvFormat < AVIF_PIXEL_FORMAT_COUNT); |
123 | 2.37k | AVIF_CHECK(image->yuvRange == AVIF_RANGE_LIMITED || image->yuvRange == AVIF_RANGE_FULL); |
124 | | |
125 | | // These matrix coefficients values are currently unsupported. Revise this list as more support is added. |
126 | | // |
127 | | // YCgCo performs limited-full range adjustment on R,G,B but the current implementation performs range adjustment |
128 | | // on Y,U,V. So YCgCo with limited range is unsupported. |
129 | 2.37k | if ((image->matrixCoefficients == 3 /* CICP reserved */) || |
130 | 2.36k | ((image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_YCGCO || image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_YCGCO_RE || |
131 | 2.23k | image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_YCGCO_RO) && |
132 | 135 | (image->yuvRange == AVIF_RANGE_LIMITED)) || |
133 | 2.36k | (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_BT2020_CL) || |
134 | 2.36k | (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_SMPTE2085) || |
135 | 2.35k | (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_CHROMA_DERIVED_CL) || |
136 | 2.35k | (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_ICTCP) || (image->matrixCoefficients >= AVIF_MATRIX_COEFFICIENTS_LAST)) { |
137 | 147 | return AVIF_FALSE; |
138 | 147 | } |
139 | | |
140 | | // Removing 400 here would break backward behavior but would respect the spec. |
141 | 2.22k | if ((image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_IDENTITY) && (image->yuvFormat != AVIF_PIXEL_FORMAT_YUV444) && |
142 | 31 | (image->yuvFormat != AVIF_PIXEL_FORMAT_YUV400)) { |
143 | 0 | return AVIF_FALSE; |
144 | 0 | } |
145 | 2.22k | avifGetPixelFormatInfo(image->yuvFormat, &info->formatInfo); |
146 | 2.22k | avifCalcYUVCoefficients(image, &info->kr, &info->kg, &info->kb); |
147 | | |
148 | 2.22k | info->channelBytes = (image->depth > 8) ? 2 : 1; |
149 | | |
150 | 2.22k | info->depth = image->depth; |
151 | 2.22k | info->range = image->yuvRange; |
152 | 2.22k | info->maxChannel = (1 << image->depth) - 1; |
153 | 2.22k | info->biasY = (info->range == AVIF_RANGE_LIMITED) ? (float)(16 << (info->depth - 8)) : 0.0f; |
154 | 2.22k | info->biasUV = (float)(1 << (info->depth - 1)); |
155 | 2.22k | info->rangeY = (float)((info->range == AVIF_RANGE_LIMITED) ? (219 << (info->depth - 8)) : info->maxChannel); |
156 | 2.22k | info->rangeUV = (float)((info->range == AVIF_RANGE_LIMITED) ? (224 << (info->depth - 8)) : info->maxChannel); |
157 | | |
158 | 2.22k | return AVIF_TRUE; |
159 | 2.22k | } |
160 | | |
161 | | static avifBool avifPrepareReformatState(const avifImage * image, const avifRGBImage * rgb, avifReformatState * state) |
162 | 2.38k | { |
163 | 2.38k | const avifBool useYCgCoRe = (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_YCGCO_RE); |
164 | 2.38k | const avifBool useYCgCoRo = (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_YCGCO_RO); |
165 | 2.38k | if (useYCgCoRe || useYCgCoRo) { |
166 | 52 | const int bitOffset = (useYCgCoRe) ? 2 : 1; |
167 | 52 | if (image->depth - bitOffset != rgb->depth) { |
168 | 12 | return AVIF_FALSE; |
169 | 12 | } |
170 | 52 | } |
171 | | |
172 | 2.37k | AVIF_CHECK(avifGetRGBColorSpaceInfo(rgb, &state->rgb)); |
173 | 2.37k | AVIF_CHECK(avifGetYUVColorSpaceInfo(image, &state->yuv)); |
174 | | |
175 | 2.22k | state->yuv.mode = AVIF_REFORMAT_MODE_YUV_COEFFICIENTS; |
176 | | |
177 | 2.22k | if (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_IDENTITY) { |
178 | 950 | state->yuv.mode = AVIF_REFORMAT_MODE_IDENTITY; |
179 | 1.27k | } else if (image->matrixCoefficients == AVIF_MATRIX_COEFFICIENTS_YCGCO) { |
180 | 95 | state->yuv.mode = AVIF_REFORMAT_MODE_YCGCO; |
181 | 1.17k | } else if (useYCgCoRe) { |
182 | 35 | state->yuv.mode = AVIF_REFORMAT_MODE_YCGCO_RE; |
183 | 1.14k | } else if (useYCgCoRo) { |
184 | 0 | state->yuv.mode = AVIF_REFORMAT_MODE_YCGCO_RO; |
185 | 0 | } |
186 | | |
187 | 2.22k | if (state->yuv.mode != AVIF_REFORMAT_MODE_YUV_COEFFICIENTS) { |
188 | 1.08k | state->yuv.kr = 0.0f; |
189 | 1.08k | state->yuv.kg = 0.0f; |
190 | 1.08k | state->yuv.kb = 0.0f; |
191 | 1.08k | } |
192 | | |
193 | 2.22k | return AVIF_TRUE; |
194 | 2.37k | } |
195 | | |
196 | | // Formulas 20-31 from https://www.itu.int/rec/T-REC-H.273-201612-S |
197 | | static int avifYUVColorSpaceInfoYToUNorm(avifYUVColorSpaceInfo * info, float v) |
198 | 0 | { |
199 | 0 | int unorm = (int)avifRoundf(v * info->rangeY + info->biasY); |
200 | 0 | return AVIF_CLAMP(unorm, 0, info->maxChannel); |
201 | 0 | } |
202 | | |
203 | | static int avifYUVColorSpaceInfoUVToUNorm(avifYUVColorSpaceInfo * info, float v) |
204 | 0 | { |
205 | 0 | int unorm; |
206 | | |
207 | | // YCgCo performs limited-full range adjustment on R,G,B but the current implementation performs range adjustment |
208 | | // on Y,U,V. So YCgCo with limited range is unsupported. |
209 | 0 | assert((info->mode != AVIF_REFORMAT_MODE_YCGCO && info->mode != AVIF_REFORMAT_MODE_YCGCO_RE && info->mode != AVIF_REFORMAT_MODE_YCGCO_RO) || |
210 | 0 | (info->range == AVIF_RANGE_FULL)); |
211 | |
|
212 | 0 | if (info->mode == AVIF_REFORMAT_MODE_IDENTITY) { |
213 | 0 | unorm = (int)avifRoundf(v * info->rangeY + info->biasY); |
214 | 0 | } else { |
215 | 0 | unorm = (int)avifRoundf(v * info->rangeUV + info->biasUV); |
216 | 0 | } |
217 | |
|
218 | 0 | return AVIF_CLAMP(unorm, 0, info->maxChannel); |
219 | 0 | } |
220 | | |
221 | | avifResult avifImageRGBToYUV(avifImage * image, const avifRGBImage * rgb) |
222 | 0 | { |
223 | 0 | if (!rgb->pixels || rgb->format == AVIF_RGB_FORMAT_RGB_565) { |
224 | 0 | return AVIF_RESULT_REFORMAT_FAILED; |
225 | 0 | } |
226 | | |
227 | 0 | avifReformatState state; |
228 | 0 | if (!avifPrepareReformatState(image, rgb, &state)) { |
229 | 0 | return AVIF_RESULT_REFORMAT_FAILED; |
230 | 0 | } |
231 | | |
232 | 0 | if (rgb->isFloat) { |
233 | 0 | return AVIF_RESULT_NOT_IMPLEMENTED; |
234 | 0 | } |
235 | | |
236 | 0 | const avifBool hasAlpha = avifRGBFormatHasAlpha(rgb->format) && !rgb->ignoreAlpha; |
237 | 0 | avifResult allocationResult = avifImageAllocatePlanes(image, hasAlpha ? AVIF_PLANES_ALL : AVIF_PLANES_YUV); |
238 | 0 | if (allocationResult != AVIF_RESULT_OK) { |
239 | 0 | return allocationResult; |
240 | 0 | } |
241 | | |
242 | 0 | avifAlphaMultiplyMode alphaMode = AVIF_ALPHA_MULTIPLY_MODE_NO_OP; |
243 | 0 | if (hasAlpha) { |
244 | 0 | if (!rgb->alphaPremultiplied && image->alphaPremultiplied) { |
245 | 0 | alphaMode = AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY; |
246 | 0 | } else if (rgb->alphaPremultiplied && !image->alphaPremultiplied) { |
247 | 0 | alphaMode = AVIF_ALPHA_MULTIPLY_MODE_UNMULTIPLY; |
248 | 0 | } |
249 | 0 | } |
250 | |
|
251 | 0 | const avifBool isGray = avifRGBFormatIsGray(rgb->format); |
252 | 0 | avifBool converted = AVIF_FALSE; |
253 | | |
254 | | // Try converting with libsharpyuv. |
255 | 0 | if (!isGray) { |
256 | 0 | if ((rgb->chromaDownsampling == AVIF_CHROMA_DOWNSAMPLING_SHARP_YUV) && (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV420)) { |
257 | 0 | const avifResult libSharpYUVResult = avifImageRGBToYUVLibSharpYUV(image, rgb, &state); |
258 | 0 | if (libSharpYUVResult != AVIF_RESULT_OK) { |
259 | | // Return the error if sharpyuv was requested but failed for any reason, including libsharpyuv not being available. |
260 | 0 | return libSharpYUVResult; |
261 | 0 | } |
262 | 0 | converted = AVIF_TRUE; |
263 | 0 | } |
264 | | |
265 | 0 | if (!converted && !rgb->avoidLibYUV && (alphaMode == AVIF_ALPHA_MULTIPLY_MODE_NO_OP)) { |
266 | 0 | avifResult libyuvResult = avifImageRGBToYUVLibYUV(image, rgb); |
267 | 0 | if (libyuvResult == AVIF_RESULT_OK) { |
268 | 0 | converted = AVIF_TRUE; |
269 | 0 | } else if (libyuvResult != AVIF_RESULT_NOT_IMPLEMENTED) { |
270 | 0 | return libyuvResult; |
271 | 0 | } |
272 | 0 | } |
273 | 0 | } |
274 | | |
275 | 0 | if (!converted && !isGray) { |
276 | 0 | const float kr = state.yuv.kr; |
277 | 0 | const float kg = state.yuv.kg; |
278 | 0 | const float kb = state.yuv.kb; |
279 | |
|
280 | 0 | struct YUVBlock yuvBlock[2][2]; |
281 | 0 | float rgbPixel[3]; |
282 | 0 | const uint32_t rgbPixelBytes = state.rgb.pixelBytes; |
283 | 0 | const uint32_t offsetBytesR = state.rgb.offsetBytesR; |
284 | 0 | const uint32_t offsetBytesG = state.rgb.offsetBytesG; |
285 | 0 | const uint32_t offsetBytesB = state.rgb.offsetBytesB; |
286 | 0 | const uint32_t offsetBytesA = state.rgb.offsetBytesA; |
287 | 0 | const size_t rgbRowBytes = rgb->rowBytes; |
288 | 0 | const float rgbMaxChannelF = state.rgb.maxChannelF; |
289 | 0 | uint8_t * yPlane = image->yuvPlanes[AVIF_CHAN_Y]; |
290 | 0 | uint8_t * uPlane = image->yuvPlanes[AVIF_CHAN_U]; |
291 | 0 | uint8_t * vPlane = image->yuvPlanes[AVIF_CHAN_V]; |
292 | 0 | const size_t yRowBytes = image->yuvRowBytes[AVIF_CHAN_Y]; |
293 | 0 | const size_t uRowBytes = image->yuvRowBytes[AVIF_CHAN_U]; |
294 | 0 | const size_t vRowBytes = image->yuvRowBytes[AVIF_CHAN_V]; |
295 | 0 | for (size_t outerJ = 0; outerJ < image->height; outerJ += 2) { |
296 | 0 | for (size_t outerI = 0; outerI < image->width; outerI += 2) { |
297 | 0 | uint32_t blockW = 2, blockH = 2; |
298 | 0 | if ((outerI + 1) >= image->width) { |
299 | 0 | blockW = 1; |
300 | 0 | } |
301 | 0 | if ((outerJ + 1) >= image->height) { |
302 | 0 | blockH = 1; |
303 | 0 | } |
304 | | |
305 | | // Convert an entire 2x2 block to YUV, and populate any fully sampled channels as we go |
306 | 0 | for (uint32_t bJ = 0; bJ < blockH; ++bJ) { |
307 | 0 | for (uint32_t bI = 0; bI < blockW; ++bI) { |
308 | 0 | const size_t i = outerI + bI; |
309 | 0 | const size_t j = outerJ + bJ; |
310 | | |
311 | | // Unpack RGB into normalized float |
312 | 0 | if (state.rgb.channelBytes > 1) { |
313 | 0 | rgbPixel[0] = *((uint16_t *)(&rgb->pixels[offsetBytesR + (i * rgbPixelBytes) + (j * rgbRowBytes)])) / |
314 | 0 | rgbMaxChannelF; |
315 | 0 | rgbPixel[1] = *((uint16_t *)(&rgb->pixels[offsetBytesG + (i * rgbPixelBytes) + (j * rgbRowBytes)])) / |
316 | 0 | rgbMaxChannelF; |
317 | 0 | rgbPixel[2] = *((uint16_t *)(&rgb->pixels[offsetBytesB + (i * rgbPixelBytes) + (j * rgbRowBytes)])) / |
318 | 0 | rgbMaxChannelF; |
319 | 0 | } else { |
320 | 0 | rgbPixel[0] = rgb->pixels[offsetBytesR + (i * rgbPixelBytes) + (j * rgbRowBytes)] / rgbMaxChannelF; |
321 | 0 | rgbPixel[1] = rgb->pixels[offsetBytesG + (i * rgbPixelBytes) + (j * rgbRowBytes)] / rgbMaxChannelF; |
322 | 0 | rgbPixel[2] = rgb->pixels[offsetBytesB + (i * rgbPixelBytes) + (j * rgbRowBytes)] / rgbMaxChannelF; |
323 | 0 | } |
324 | |
|
325 | 0 | if (alphaMode != AVIF_ALPHA_MULTIPLY_MODE_NO_OP) { |
326 | 0 | float a; |
327 | 0 | if (state.rgb.channelBytes > 1) { |
328 | 0 | a = *((uint16_t *)(&rgb->pixels[offsetBytesA + (i * rgbPixelBytes) + (j * rgbRowBytes)])) / rgbMaxChannelF; |
329 | 0 | } else { |
330 | 0 | a = rgb->pixels[offsetBytesA + (i * rgbPixelBytes) + (j * rgbRowBytes)] / rgbMaxChannelF; |
331 | 0 | } |
332 | |
|
333 | 0 | if (alphaMode == AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY) { |
334 | 0 | if (a == 0) { |
335 | 0 | rgbPixel[0] = 0; |
336 | 0 | rgbPixel[1] = 0; |
337 | 0 | rgbPixel[2] = 0; |
338 | 0 | } else if (a < 1.0f) { |
339 | 0 | rgbPixel[0] *= a; |
340 | 0 | rgbPixel[1] *= a; |
341 | 0 | rgbPixel[2] *= a; |
342 | 0 | } |
343 | 0 | } else { |
344 | | // alphaMode == AVIF_ALPHA_MULTIPLY_MODE_UNMULTIPLY |
345 | 0 | if (a == 0) { |
346 | 0 | rgbPixel[0] = 0; |
347 | 0 | rgbPixel[1] = 0; |
348 | 0 | rgbPixel[2] = 0; |
349 | 0 | } else if (a < 1.0f) { |
350 | 0 | rgbPixel[0] /= a; |
351 | 0 | rgbPixel[1] /= a; |
352 | 0 | rgbPixel[2] /= a; |
353 | 0 | rgbPixel[0] = AVIF_MIN(rgbPixel[0], 1.0f); |
354 | 0 | rgbPixel[1] = AVIF_MIN(rgbPixel[1], 1.0f); |
355 | 0 | rgbPixel[2] = AVIF_MIN(rgbPixel[2], 1.0f); |
356 | 0 | } |
357 | 0 | } |
358 | 0 | } |
359 | | |
360 | | // RGB -> YUV conversion |
361 | 0 | if (state.yuv.mode == AVIF_REFORMAT_MODE_IDENTITY) { |
362 | | // Formulas 41,42,43 from https://www.itu.int/rec/T-REC-H.273-201612-S |
363 | 0 | yuvBlock[bI][bJ].y = rgbPixel[1]; // G |
364 | 0 | yuvBlock[bI][bJ].u = rgbPixel[2]; // B |
365 | 0 | yuvBlock[bI][bJ].v = rgbPixel[0]; // R |
366 | 0 | } else if (state.yuv.mode == AVIF_REFORMAT_MODE_YCGCO) { |
367 | | // Formulas 44,45,46 from https://www.itu.int/rec/T-REC-H.273-201612-S |
368 | 0 | yuvBlock[bI][bJ].y = 0.5f * rgbPixel[1] + 0.25f * (rgbPixel[0] + rgbPixel[2]); |
369 | 0 | yuvBlock[bI][bJ].u = 0.5f * rgbPixel[1] - 0.25f * (rgbPixel[0] + rgbPixel[2]); |
370 | 0 | yuvBlock[bI][bJ].v = 0.5f * (rgbPixel[0] - rgbPixel[2]); |
371 | 0 | } else if (state.yuv.mode == AVIF_REFORMAT_MODE_YCGCO_RE || state.yuv.mode == AVIF_REFORMAT_MODE_YCGCO_RO) { |
372 | | // Formulas 58,59,60,61 from https://www.itu.int/rec/T-REC-H.273-202407-P |
373 | 0 | const int R = (int)avifRoundf(AVIF_CLAMP(rgbPixel[0] * rgbMaxChannelF, 0.0f, rgbMaxChannelF)); |
374 | 0 | const int G = (int)avifRoundf(AVIF_CLAMP(rgbPixel[1] * rgbMaxChannelF, 0.0f, rgbMaxChannelF)); |
375 | 0 | const int B = (int)avifRoundf(AVIF_CLAMP(rgbPixel[2] * rgbMaxChannelF, 0.0f, rgbMaxChannelF)); |
376 | 0 | const int Co = R - B; |
377 | 0 | const int t = B + (Co >> 1); |
378 | 0 | const int Cg = G - t; |
379 | 0 | yuvBlock[bI][bJ].y = (t + (Cg >> 1)) / state.yuv.rangeY; |
380 | 0 | yuvBlock[bI][bJ].u = Cg / state.yuv.rangeUV; |
381 | 0 | yuvBlock[bI][bJ].v = Co / state.yuv.rangeUV; |
382 | 0 | } else { |
383 | 0 | float Y = (kr * rgbPixel[0]) + (kg * rgbPixel[1]) + (kb * rgbPixel[2]); |
384 | 0 | yuvBlock[bI][bJ].y = Y; |
385 | 0 | yuvBlock[bI][bJ].u = (rgbPixel[2] - Y) / (2 * (1 - kb)); |
386 | 0 | yuvBlock[bI][bJ].v = (rgbPixel[0] - Y) / (2 * (1 - kr)); |
387 | 0 | } |
388 | |
|
389 | 0 | if (state.yuv.channelBytes > 1) { |
390 | 0 | uint16_t * pY = (uint16_t *)&yPlane[(i * 2) + (j * yRowBytes)]; |
391 | 0 | *pY = (uint16_t)avifYUVColorSpaceInfoYToUNorm(&state.yuv, yuvBlock[bI][bJ].y); |
392 | 0 | if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV444) { |
393 | | // YUV444, full chroma |
394 | 0 | uint16_t * pU = (uint16_t *)&uPlane[(i * 2) + (j * uRowBytes)]; |
395 | 0 | *pU = (uint16_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, yuvBlock[bI][bJ].u); |
396 | 0 | uint16_t * pV = (uint16_t *)&vPlane[(i * 2) + (j * vRowBytes)]; |
397 | 0 | *pV = (uint16_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, yuvBlock[bI][bJ].v); |
398 | 0 | } |
399 | 0 | } else { |
400 | 0 | yPlane[i + (j * yRowBytes)] = (uint8_t)avifYUVColorSpaceInfoYToUNorm(&state.yuv, yuvBlock[bI][bJ].y); |
401 | 0 | if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV444) { |
402 | | // YUV444, full chroma |
403 | 0 | uPlane[i + (j * uRowBytes)] = (uint8_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, yuvBlock[bI][bJ].u); |
404 | 0 | vPlane[i + (j * vRowBytes)] = (uint8_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, yuvBlock[bI][bJ].v); |
405 | 0 | } |
406 | 0 | } |
407 | 0 | } |
408 | 0 | } |
409 | | |
410 | | // Populate any subsampled channels with averages from the 2x2 block |
411 | 0 | if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV400) { |
412 | | // Do nothing on chroma planes. |
413 | 0 | } else if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV420) { |
414 | | // YUV420, average 4 samples (2x2) |
415 | |
|
416 | 0 | float sumU = 0.0f; |
417 | 0 | float sumV = 0.0f; |
418 | 0 | for (uint32_t bJ = 0; bJ < blockH; ++bJ) { |
419 | 0 | for (uint32_t bI = 0; bI < blockW; ++bI) { |
420 | 0 | sumU += yuvBlock[bI][bJ].u; |
421 | 0 | sumV += yuvBlock[bI][bJ].v; |
422 | 0 | } |
423 | 0 | } |
424 | 0 | float totalSamples = (float)(blockW * blockH); |
425 | 0 | float avgU = sumU / totalSamples; |
426 | 0 | float avgV = sumV / totalSamples; |
427 | |
|
428 | 0 | const int chromaShiftX = 1; |
429 | 0 | const int chromaShiftY = 1; |
430 | 0 | size_t uvI = outerI >> chromaShiftX; |
431 | 0 | size_t uvJ = outerJ >> chromaShiftY; |
432 | 0 | if (state.yuv.channelBytes > 1) { |
433 | 0 | uint16_t * pU = (uint16_t *)&uPlane[(uvI * 2) + (uvJ * uRowBytes)]; |
434 | 0 | *pU = (uint16_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgU); |
435 | 0 | uint16_t * pV = (uint16_t *)&vPlane[(uvI * 2) + (uvJ * vRowBytes)]; |
436 | 0 | *pV = (uint16_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgV); |
437 | 0 | } else { |
438 | 0 | uPlane[uvI + (uvJ * uRowBytes)] = (uint8_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgU); |
439 | 0 | vPlane[uvI + (uvJ * vRowBytes)] = (uint8_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgV); |
440 | 0 | } |
441 | 0 | } else if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV422) { |
442 | | // YUV422, average 2 samples (1x2), twice |
443 | |
|
444 | 0 | for (uint32_t bJ = 0; bJ < blockH; ++bJ) { |
445 | 0 | float sumU = 0.0f; |
446 | 0 | float sumV = 0.0f; |
447 | 0 | for (uint32_t bI = 0; bI < blockW; ++bI) { |
448 | 0 | sumU += yuvBlock[bI][bJ].u; |
449 | 0 | sumV += yuvBlock[bI][bJ].v; |
450 | 0 | } |
451 | 0 | float totalSamples = (float)blockW; |
452 | 0 | float avgU = sumU / totalSamples; |
453 | 0 | float avgV = sumV / totalSamples; |
454 | |
|
455 | 0 | const int chromaShiftX = 1; |
456 | 0 | size_t uvI = outerI >> chromaShiftX; |
457 | 0 | size_t uvJ = outerJ + bJ; |
458 | 0 | if (state.yuv.channelBytes > 1) { |
459 | 0 | uint16_t * pU = (uint16_t *)&uPlane[(uvI * 2) + (uvJ * uRowBytes)]; |
460 | 0 | *pU = (uint16_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgU); |
461 | 0 | uint16_t * pV = (uint16_t *)&vPlane[(uvI * 2) + (uvJ * vRowBytes)]; |
462 | 0 | *pV = (uint16_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgV); |
463 | 0 | } else { |
464 | 0 | uPlane[uvI + (uvJ * uRowBytes)] = (uint8_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgU); |
465 | 0 | vPlane[uvI + (uvJ * vRowBytes)] = (uint8_t)avifYUVColorSpaceInfoUVToUNorm(&state.yuv, avgV); |
466 | 0 | } |
467 | 0 | } |
468 | 0 | } |
469 | 0 | } |
470 | 0 | } |
471 | 0 | } else if (!converted && isGray) { |
472 | 0 | const uint32_t grayPixelBytes = state.rgb.pixelBytes; |
473 | 0 | const uint32_t offsetBytesGray = state.rgb.offsetBytesGray; |
474 | 0 | const uint32_t offsetBytesA = state.rgb.offsetBytesA; |
475 | 0 | const size_t grayRowBytes = rgb->rowBytes; |
476 | 0 | const float grayMaxChannelF = state.rgb.maxChannelF; |
477 | 0 | uint8_t * yPlane = image->yuvPlanes[AVIF_CHAN_Y]; |
478 | 0 | const size_t yRowBytes = image->yuvRowBytes[AVIF_CHAN_Y]; |
479 | 0 | for (size_t j = 0; j < image->height; ++j) { |
480 | 0 | for (size_t i = 0; i < image->width; ++i) { |
481 | 0 | float g; |
482 | 0 | if (state.rgb.channelBytes > 1) { |
483 | 0 | g = *(uint16_t *)&rgb->pixels[offsetBytesGray + i * grayPixelBytes + (j * grayRowBytes)] / grayMaxChannelF; |
484 | 0 | } else { |
485 | 0 | g = rgb->pixels[offsetBytesGray + i * grayPixelBytes + (j * grayRowBytes)] / grayMaxChannelF; |
486 | 0 | } |
487 | 0 | if (alphaMode != AVIF_ALPHA_MULTIPLY_MODE_NO_OP) { |
488 | 0 | float a; |
489 | 0 | if (state.rgb.channelBytes > 1) { |
490 | 0 | a = *((uint16_t *)(&rgb->pixels[offsetBytesA + (i * grayPixelBytes) + (j * grayRowBytes)])) / grayMaxChannelF; |
491 | 0 | } else { |
492 | 0 | a = rgb->pixels[offsetBytesA + (i * grayPixelBytes) + (j * grayRowBytes)] / grayMaxChannelF; |
493 | 0 | } |
494 | |
|
495 | 0 | if (alphaMode == AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY) { |
496 | 0 | if (a == 0) { |
497 | 0 | g = 0; |
498 | 0 | } else if (a < 1.0f) { |
499 | 0 | g *= a; |
500 | 0 | } |
501 | 0 | } else { |
502 | | // alphaMode == AVIF_ALPHA_MULTIPLY_MODE_UNMULTIPLY |
503 | 0 | if (a == 0) { |
504 | 0 | g = 0; |
505 | 0 | } else if (a < 1.0f) { |
506 | 0 | g /= a; |
507 | 0 | g = AVIF_MIN(g, 1.0f); |
508 | 0 | } |
509 | 0 | } |
510 | 0 | } |
511 | 0 | int gInt = avifYUVColorSpaceInfoYToUNorm(&state.yuv, g); |
512 | 0 | if (state.yuv.channelBytes > 1) { |
513 | 0 | uint16_t * pY = (uint16_t *)&yPlane[(i * 2) + j * yRowBytes]; |
514 | 0 | *pY = (uint16_t)gInt; |
515 | 0 | } else { |
516 | 0 | yPlane[i + (j * yRowBytes)] = (uint8_t)gInt; |
517 | 0 | } |
518 | 0 | } |
519 | 0 | } |
520 | | // Set the chroma planes, if any, to the half value. |
521 | 0 | avifPixelFormatInfo info; |
522 | 0 | avifGetPixelFormatInfo(image->yuvFormat, &info); |
523 | 0 | const uint32_t shiftedH = (uint32_t)(((uint64_t)image->height + info.chromaShiftY) >> info.chromaShiftY); |
524 | 0 | const int half = 1 << (image->depth - 1); |
525 | 0 | if (image->yuvPlanes[AVIF_CHAN_U]) { |
526 | 0 | uint8_t * uPlane = image->yuvPlanes[AVIF_CHAN_U]; |
527 | 0 | const size_t uRowBytes = image->yuvRowBytes[AVIF_CHAN_U]; |
528 | 0 | if (state.yuv.channelBytes > 1) { |
529 | 0 | avifMemset16(uPlane, half, shiftedH * uRowBytes / 2); |
530 | 0 | } else { |
531 | 0 | memset(uPlane, half, shiftedH * uRowBytes); |
532 | 0 | } |
533 | 0 | } |
534 | 0 | if (image->yuvPlanes[AVIF_CHAN_V]) { |
535 | 0 | uint8_t * vPlane = image->yuvPlanes[AVIF_CHAN_V]; |
536 | 0 | const size_t vRowBytes = image->yuvRowBytes[AVIF_CHAN_V]; |
537 | 0 | if (state.yuv.channelBytes > 1) { |
538 | 0 | avifMemset16(vPlane, half, shiftedH * vRowBytes / 2); |
539 | 0 | } else { |
540 | 0 | memset(vPlane, half, shiftedH * vRowBytes); |
541 | 0 | } |
542 | 0 | } |
543 | 0 | } |
544 | |
|
545 | 0 | if (image->alphaPlane && image->alphaRowBytes) { |
546 | 0 | avifAlphaParams params; |
547 | |
|
548 | 0 | params.width = image->width; |
549 | 0 | params.height = image->height; |
550 | 0 | params.dstDepth = image->depth; |
551 | 0 | params.dstPlane = image->alphaPlane; |
552 | 0 | params.dstRowBytes = image->alphaRowBytes; |
553 | 0 | params.dstOffsetBytes = 0; |
554 | 0 | params.dstPixelBytes = state.yuv.channelBytes; |
555 | |
|
556 | 0 | if (avifRGBFormatHasAlpha(rgb->format) && !rgb->ignoreAlpha) { |
557 | 0 | params.srcDepth = rgb->depth; |
558 | 0 | params.srcPlane = rgb->pixels; |
559 | 0 | params.srcRowBytes = rgb->rowBytes; |
560 | 0 | params.srcOffsetBytes = state.rgb.offsetBytesA; |
561 | 0 | params.srcPixelBytes = state.rgb.pixelBytes; |
562 | |
|
563 | 0 | avifReformatAlpha(¶ms); |
564 | 0 | } else { |
565 | | // libyuv does not fill alpha when converting from RGB to YUV so |
566 | | // fill it regardless of the value of convertedWithLibYUV. |
567 | 0 | avifFillAlpha(¶ms); |
568 | 0 | } |
569 | 0 | } |
570 | 0 | return AVIF_RESULT_OK; |
571 | 0 | } |
572 | | |
573 | | // Allocates and fills look-up tables for going from YUV limited/full unorm -> full range RGB FP32. |
574 | | // Review this when implementing YCgCo limited range support. |
575 | | static avifBool avifCreateYUVToRGBLookUpTables(float ** unormFloatTableY, float ** unormFloatTableUV, uint32_t depth, const avifReformatState * state) |
576 | 6.30k | { |
577 | 6.30k | const size_t cpCount = (size_t)1 << depth; |
578 | | |
579 | 6.30k | assert(unormFloatTableY); |
580 | 6.30k | *unormFloatTableY = (float *)avifAlloc(cpCount * sizeof(float)); |
581 | 6.30k | AVIF_CHECK(*unormFloatTableY); |
582 | 7.61M | for (uint32_t cp = 0; cp < cpCount; ++cp) { |
583 | 7.61M | (*unormFloatTableY)[cp] = ((float)cp - state->yuv.biasY) / state->yuv.rangeY; |
584 | 7.61M | } |
585 | | |
586 | 6.30k | if (unormFloatTableUV) { |
587 | 5.02k | if (state->yuv.mode == AVIF_REFORMAT_MODE_IDENTITY) { |
588 | | // Just reuse the luma table since the chroma values are the same. |
589 | 384 | *unormFloatTableUV = *unormFloatTableY; |
590 | 4.64k | } else { |
591 | 4.64k | *unormFloatTableUV = (float *)avifAlloc(cpCount * sizeof(float)); |
592 | 4.64k | if (!*unormFloatTableUV) { |
593 | 0 | avifFree(*unormFloatTableY); |
594 | 0 | *unormFloatTableY = NULL; |
595 | 0 | return AVIF_FALSE; |
596 | 0 | } |
597 | 6.46M | for (uint32_t cp = 0; cp < cpCount; ++cp) { |
598 | 6.45M | (*unormFloatTableUV)[cp] = ((float)cp - state->yuv.biasUV) / state->yuv.rangeUV; |
599 | 6.45M | } |
600 | 4.64k | } |
601 | 5.02k | } |
602 | 6.30k | return AVIF_TRUE; |
603 | 6.30k | } |
604 | | |
605 | | // Frees look-up tables allocated with avifCreateYUVToRGBLookUpTables(). |
606 | | static void avifFreeYUVToRGBLookUpTables(float ** unormFloatTableY, float ** unormFloatTableUV) |
607 | 6.30k | { |
608 | 6.30k | if (unormFloatTableUV) { |
609 | 5.02k | if (*unormFloatTableUV != *unormFloatTableY) { |
610 | 4.64k | avifFree(*unormFloatTableUV); |
611 | 4.64k | } |
612 | 5.02k | *unormFloatTableUV = NULL; |
613 | 5.02k | } |
614 | | |
615 | 6.30k | avifFree(*unormFloatTableY); |
616 | 6.30k | *unormFloatTableY = NULL; |
617 | 6.30k | } |
618 | | |
619 | 0 | #define RGB565(R, G, B) ((uint16_t)(((B) >> 3) | (((G) >> 2) << 5) | (((R) >> 3) << 11))) |
620 | | |
621 | | static void avifStoreRGB8Pixel(avifRGBFormat format, uint8_t R, uint8_t G, uint8_t B, uint8_t * ptrR, uint8_t * ptrG, uint8_t * ptrB) |
622 | 819M | { |
623 | 819M | if (format == AVIF_RGB_FORMAT_RGB_565) { |
624 | | // References for RGB565 color conversion: |
625 | | // * https://docs.microsoft.com/en-us/windows/win32/directshow/working-with-16-bit-rgb |
626 | | // * https://chromium.googlesource.com/libyuv/libyuv/+/9892d70c965678381d2a70a1c9002d1cf136ee78/source/row_common.cc#2362 |
627 | 0 | *(uint16_t *)ptrR = RGB565(R, G, B); |
628 | 0 | return; |
629 | 0 | } |
630 | 819M | *ptrR = R; |
631 | 819M | *ptrG = G; |
632 | 819M | *ptrB = B; |
633 | 819M | } |
634 | | |
635 | | static void avifGetRGB565(const uint8_t * ptrR, uint8_t * R, uint8_t * G, uint8_t * B) |
636 | 0 | { |
637 | | // References for RGB565 color conversion: |
638 | | // * https://docs.microsoft.com/en-us/windows/win32/directshow/working-with-16-bit-rgb |
639 | | // * https://chromium.googlesource.com/libyuv/libyuv/+/331c361581896292fb46c8c6905e41262b7ca95f/source/row_common.cc#185 |
640 | 0 | const uint16_t rgb656 = ((const uint16_t *)ptrR)[0]; |
641 | 0 | const uint16_t r5 = (rgb656 & 0xF800) >> 11; |
642 | 0 | const uint16_t g6 = (rgb656 & 0x07E0) >> 5; |
643 | 0 | const uint16_t b5 = (rgb656 & 0x001F); |
644 | 0 | *R = (uint8_t)((r5 << 3) | (r5 >> 2)); |
645 | 0 | *G = (uint8_t)((g6 << 2) | (g6 >> 4)); |
646 | 0 | *B = (uint8_t)((b5 << 3) | (b5 >> 2)); |
647 | 0 | } |
648 | | |
649 | | // Note: This function handles alpha (un)multiply. |
650 | | static avifResult avifImageYUVAnyToRGBAnySlow(const avifImage * image, |
651 | | avifRGBImage * rgb, |
652 | | const avifReformatState * state, |
653 | | avifAlphaMultiplyMode alphaMultiplyMode) |
654 | 2.13k | { |
655 | | // Aliases for some state |
656 | 2.13k | const float kr = state->yuv.kr; |
657 | 2.13k | const float kg = state->yuv.kg; |
658 | 2.13k | const float kb = state->yuv.kb; |
659 | 2.13k | float * unormFloatTableY = NULL; |
660 | 2.13k | float * unormFloatTableUV = NULL; |
661 | 2.13k | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
662 | 2.13k | const uint32_t yuvChannelBytes = state->yuv.channelBytes; |
663 | 2.13k | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
664 | | |
665 | | // Aliases for plane data |
666 | 2.13k | const uint8_t * yPlane = image->yuvPlanes[AVIF_CHAN_Y]; |
667 | 2.13k | const uint8_t * uPlane = image->yuvPlanes[AVIF_CHAN_U]; |
668 | 2.13k | const uint8_t * vPlane = image->yuvPlanes[AVIF_CHAN_V]; |
669 | 2.13k | const uint8_t * aPlane = image->alphaPlane; |
670 | 2.13k | const uint32_t yRowBytes = image->yuvRowBytes[AVIF_CHAN_Y]; |
671 | 2.13k | const uint32_t uRowBytes = image->yuvRowBytes[AVIF_CHAN_U]; |
672 | 2.13k | const uint32_t vRowBytes = image->yuvRowBytes[AVIF_CHAN_V]; |
673 | 2.13k | const uint32_t aRowBytes = image->alphaRowBytes; |
674 | | |
675 | | // Various observations and limits |
676 | 2.13k | const avifBool yuvHasColor = (uPlane && vPlane && (image->yuvFormat != AVIF_PIXEL_FORMAT_YUV400)); |
677 | 2.13k | const avifBool rgbHasColor = !avifRGBFormatIsGray(rgb->format); |
678 | 2.13k | const uint16_t yuvMaxChannel = (uint16_t)state->yuv.maxChannel; |
679 | 2.13k | const float rgbMaxChannelF = state->rgb.maxChannelF; |
680 | | |
681 | | // If toRGBAlphaMode is active (not no-op), assert that the alpha plane is present. The end of |
682 | | // the avifPrepareReformatState() function should ensure this, but this assert makes it clear |
683 | | // to clang's analyzer. |
684 | 2.13k | assert((alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_NO_OP) || aPlane); |
685 | | |
686 | 3.20M | for (uint32_t j = 0; j < image->height; ++j) { |
687 | | // uvJ is used only when yuvHasColor is true. |
688 | 3.20M | const uint32_t uvJ = yuvHasColor ? (j >> state->yuv.formatInfo.chromaShiftY) : 0; |
689 | 3.20M | const uint8_t * ptrY8 = &yPlane[j * yRowBytes]; |
690 | 3.20M | const uint8_t * ptrU8 = uPlane ? &uPlane[(uvJ * uRowBytes)] : NULL; |
691 | 3.20M | const uint8_t * ptrV8 = vPlane ? &vPlane[(uvJ * vRowBytes)] : NULL; |
692 | 3.20M | const uint8_t * ptrA8 = aPlane ? &aPlane[j * aRowBytes] : NULL; |
693 | 3.20M | const uint16_t * ptrY16 = (const uint16_t *)ptrY8; |
694 | 3.20M | const uint16_t * ptrU16 = (const uint16_t *)ptrU8; |
695 | 3.20M | const uint16_t * ptrV16 = (const uint16_t *)ptrV8; |
696 | 3.20M | const uint16_t * ptrA16 = (const uint16_t *)ptrA8; |
697 | | |
698 | 3.20M | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + ((size_t)j * rgb->rowBytes)]; |
699 | 3.20M | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + ((size_t)j * rgb->rowBytes)]; |
700 | 3.20M | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + ((size_t)j * rgb->rowBytes)]; |
701 | 3.20M | uint8_t * ptrGray = &rgb->pixels[state->rgb.offsetBytesGray + ((size_t)j * rgb->rowBytes)]; |
702 | | |
703 | 436M | for (uint32_t i = 0; i < image->width; ++i) { |
704 | 433M | float Y, Cb = 0.5f, Cr = 0.5f; |
705 | | |
706 | | // Calculate Y |
707 | 433M | uint16_t unormY; |
708 | 433M | if (image->depth == 8) { |
709 | 205M | unormY = ptrY8[i]; |
710 | 227M | } else { |
711 | | // clamp incoming data to protect against bad LUT lookups |
712 | 227M | unormY = AVIF_MIN(ptrY16[i], yuvMaxChannel); |
713 | 227M | } |
714 | 433M | Y = unormFloatTableY[unormY]; |
715 | | |
716 | | // Calculate Cb and Cr |
717 | 433M | if (yuvHasColor) { |
718 | 390M | const uint32_t uvI = i >> state->yuv.formatInfo.chromaShiftX; |
719 | 390M | if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV444) { |
720 | 26.0M | uint16_t unormU, unormV; |
721 | | |
722 | 26.0M | if (image->depth == 8) { |
723 | 10.8M | unormU = ptrU8[uvI]; |
724 | 10.8M | unormV = ptrV8[uvI]; |
725 | 15.2M | } else { |
726 | | // clamp incoming data to protect against bad LUT lookups |
727 | 15.2M | unormU = AVIF_MIN(ptrU16[uvI], yuvMaxChannel); |
728 | 15.2M | unormV = AVIF_MIN(ptrV16[uvI], yuvMaxChannel); |
729 | 15.2M | } |
730 | | |
731 | 26.0M | Cb = unormFloatTableUV[unormU]; |
732 | 26.0M | Cr = unormFloatTableUV[unormV]; |
733 | 364M | } else { |
734 | | // Upsample to 444: |
735 | | // |
736 | | // * * * * |
737 | | // A B |
738 | | // * 1 2 * |
739 | | // |
740 | | // * 3 4 * |
741 | | // C D |
742 | | // * * * * |
743 | | // |
744 | | // When converting from YUV420 to RGB, for any given "high-resolution" RGB |
745 | | // coordinate (1,2,3,4,*), there are up to four "low-resolution" UV samples |
746 | | // (A,B,C,D) that are "nearest" to the pixel. For RGB pixel #1, A is the closest |
747 | | // UV sample, B and C are "adjacent" to it on the same row and column, and D is |
748 | | // the diagonal. For RGB pixel 3, C is the closest UV sample, A and D are |
749 | | // adjacent, and B is the diagonal. Sometimes the adjacent pixel on the same row |
750 | | // is to the left or right, and sometimes the adjacent pixel on the same column |
751 | | // is up or down. For any edge or corner, there might only be only one or two |
752 | | // samples nearby, so they'll be duplicated. |
753 | | // |
754 | | // The following code attempts to find all four nearest UV samples and put them |
755 | | // in the following unormU and unormV grid as follows: |
756 | | // |
757 | | // unorm[0][0] = closest ( weights: bilinear: 9/16, nearest: 1 ) |
758 | | // unorm[1][0] = adjacent col ( weights: bilinear: 3/16, nearest: 0 ) |
759 | | // unorm[0][1] = adjacent row ( weights: bilinear: 3/16, nearest: 0 ) |
760 | | // unorm[1][1] = diagonal ( weights: bilinear: 1/16, nearest: 0 ) |
761 | | // |
762 | | // It then weights them according to the requested upsampling set in avifRGBImage. |
763 | | |
764 | 364M | uint16_t unormU[2][2], unormV[2][2]; |
765 | | |
766 | | // How many bytes to add to a uint8_t pointer index to get to the adjacent (lesser) sample in a given direction |
767 | 364M | int uAdjCol, vAdjCol, uAdjRow, vAdjRow; |
768 | 364M | if ((i == 0) || ((i == (image->width - 1)) && ((i % 2) != 0))) { |
769 | 3.31M | uAdjCol = 0; |
770 | 3.31M | vAdjCol = 0; |
771 | 361M | } else { |
772 | 361M | if ((i % 2) != 0) { |
773 | 186M | uAdjCol = yuvChannelBytes; |
774 | 186M | vAdjCol = yuvChannelBytes; |
775 | 186M | } else { |
776 | 175M | uAdjCol = -1 * yuvChannelBytes; |
777 | 175M | vAdjCol = -1 * yuvChannelBytes; |
778 | 175M | } |
779 | 361M | } |
780 | | |
781 | | // For YUV422, uvJ will always be a fresh value (always corresponds to j), so |
782 | | // we'll simply duplicate the sample as if we were on the top or bottom row and |
783 | | // it'll behave as plain old linear (1D) upsampling, which is all we want. |
784 | 364M | if ((j == 0) || ((j == (image->height - 1)) && ((j % 2) != 0)) || (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV422)) { |
785 | 29.8M | uAdjRow = 0; |
786 | 29.8M | vAdjRow = 0; |
787 | 334M | } else { |
788 | 334M | if ((j % 2) != 0) { |
789 | 166M | uAdjRow = (int)uRowBytes; |
790 | 166M | vAdjRow = (int)vRowBytes; |
791 | 167M | } else { |
792 | 167M | uAdjRow = -1 * (int)uRowBytes; |
793 | 167M | vAdjRow = -1 * (int)vRowBytes; |
794 | 167M | } |
795 | 334M | } |
796 | | |
797 | 364M | if (image->depth == 8) { |
798 | 191M | unormU[0][0] = uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes)]; |
799 | 191M | unormV[0][0] = vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes)]; |
800 | 191M | unormU[1][0] = uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes) + uAdjCol]; |
801 | 191M | unormV[1][0] = vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes) + vAdjCol]; |
802 | 191M | unormU[0][1] = uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes) + uAdjRow]; |
803 | 191M | unormV[0][1] = vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes) + vAdjRow]; |
804 | 191M | unormU[1][1] = uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes) + uAdjCol + uAdjRow]; |
805 | 191M | unormV[1][1] = vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes) + vAdjCol + vAdjRow]; |
806 | 191M | } else { |
807 | 172M | unormU[0][0] = *((const uint16_t *)&uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes)]); |
808 | 172M | unormV[0][0] = *((const uint16_t *)&vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes)]); |
809 | 172M | unormU[1][0] = *((const uint16_t *)&uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes) + uAdjCol]); |
810 | 172M | unormV[1][0] = *((const uint16_t *)&vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes) + vAdjCol]); |
811 | 172M | unormU[0][1] = *((const uint16_t *)&uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes) + uAdjRow]); |
812 | 172M | unormV[0][1] = *((const uint16_t *)&vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes) + vAdjRow]); |
813 | 172M | unormU[1][1] = *((const uint16_t *)&uPlane[(uvJ * uRowBytes) + (uvI * yuvChannelBytes) + uAdjCol + uAdjRow]); |
814 | 172M | unormV[1][1] = *((const uint16_t *)&vPlane[(uvJ * vRowBytes) + (uvI * yuvChannelBytes) + vAdjCol + vAdjRow]); |
815 | | |
816 | | // clamp incoming data to protect against bad LUT lookups |
817 | 505M | for (int bJ = 0; bJ < 2; ++bJ) { |
818 | 1.00G | for (int bI = 0; bI < 2; ++bI) { |
819 | 668M | unormU[bI][bJ] = AVIF_MIN(unormU[bI][bJ], yuvMaxChannel); |
820 | 668M | unormV[bI][bJ] = AVIF_MIN(unormV[bI][bJ], yuvMaxChannel); |
821 | 668M | } |
822 | 333M | } |
823 | 172M | } |
824 | | |
825 | 364M | if ((rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_FASTEST) || |
826 | 360M | (rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_NEAREST)) { |
827 | | // Nearest neighbor; ignore all UVs but the closest one |
828 | 0 | Cb = unormFloatTableUV[unormU[0][0]]; |
829 | 0 | Cr = unormFloatTableUV[unormV[0][0]]; |
830 | 364M | } else { |
831 | | // Bilinear filtering with weights |
832 | 364M | Cb = (unormFloatTableUV[unormU[0][0]] * (9.0f / 16.0f)) + (unormFloatTableUV[unormU[1][0]] * (3.0f / 16.0f)) + |
833 | 364M | (unormFloatTableUV[unormU[0][1]] * (3.0f / 16.0f)) + (unormFloatTableUV[unormU[1][1]] * (1.0f / 16.0f)); |
834 | 364M | Cr = (unormFloatTableUV[unormV[0][0]] * (9.0f / 16.0f)) + (unormFloatTableUV[unormV[1][0]] * (3.0f / 16.0f)) + |
835 | 364M | (unormFloatTableUV[unormV[0][1]] * (3.0f / 16.0f)) + (unormFloatTableUV[unormV[1][1]] * (1.0f / 16.0f)); |
836 | 364M | } |
837 | 364M | } |
838 | 390M | } |
839 | | |
840 | 433M | float Rc = 0.0f, Gc = 0.0f, Bc = 0.0f, grayc = 0.0f; |
841 | 433M | if (rgbHasColor) { |
842 | 432M | float R, G, B; |
843 | 432M | if (yuvHasColor) { |
844 | 387M | if (state->yuv.mode == AVIF_REFORMAT_MODE_IDENTITY) { |
845 | | // Identity (GBR): Formulas 41,42,43 from |
846 | | // https://www.itu.int/rec/T-REC-H.273-201612-S |
847 | 12.6M | G = Y; |
848 | 12.6M | B = Cb; |
849 | 12.6M | R = Cr; |
850 | 374M | } else if (state->yuv.mode == AVIF_REFORMAT_MODE_YCGCO) { |
851 | | // YCgCo: Formulas 47,48,49,50 from |
852 | | // https://www.itu.int/rec/T-REC-H.273-201612-S |
853 | 10.5M | const float t = Y - Cb; |
854 | 10.5M | G = Y + Cb; |
855 | 10.5M | B = t - Cr; |
856 | 10.5M | R = t + Cr; |
857 | 364M | } else if ((state->yuv.mode == AVIF_REFORMAT_MODE_YCGCO_RE) || (state->yuv.mode == AVIF_REFORMAT_MODE_YCGCO_RO)) { |
858 | | // YCgCoRe/YCgCoRo: Formulas 62,63,64,65 from |
859 | | // https://www.itu.int/rec/T-REC-H.273-202407-P |
860 | 2.99M | const int YY = unormY; |
861 | 2.99M | const int Cg = (int)avifRoundf(Cb * yuvMaxChannel); |
862 | 2.99M | const int Co = (int)avifRoundf(Cr * yuvMaxChannel); |
863 | 2.99M | const int t = YY - (Cg >> 1); |
864 | 2.99M | G = (float)AVIF_CLAMP(t + Cg, 0, state->rgb.maxChannel); |
865 | 2.99M | B = (float)AVIF_CLAMP(t - (Co >> 1), 0, state->rgb.maxChannel); |
866 | 2.99M | R = (float)AVIF_CLAMP(B + Co, 0, state->rgb.maxChannel); |
867 | 2.99M | G /= rgbMaxChannelF; |
868 | 2.99M | B /= rgbMaxChannelF; |
869 | 2.99M | R /= rgbMaxChannelF; |
870 | 361M | } else { |
871 | | // Normal YUV |
872 | 361M | R = Y + (2 * (1 - kr)) * Cr; |
873 | 361M | B = Y + (2 * (1 - kb)) * Cb; |
874 | 361M | G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
875 | 361M | } |
876 | 387M | } else { |
877 | | // Monochrome: just populate all channels with luma (state->yuv.mode |
878 | | // is irrelevant) |
879 | 45.3M | R = Y; |
880 | 45.3M | G = Y; |
881 | 45.3M | B = Y; |
882 | 45.3M | } |
883 | 432M | Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
884 | 432M | Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
885 | 432M | Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
886 | 432M | } else { |
887 | | // Monochrome: gray is luma |
888 | 564k | float gray = Y; |
889 | 564k | grayc = AVIF_CLAMP(gray, 0.0f, 1.0f); |
890 | 564k | } |
891 | | |
892 | 433M | if (alphaMultiplyMode != AVIF_ALPHA_MULTIPLY_MODE_NO_OP) { |
893 | | // Calculate A |
894 | 0 | uint16_t unormA; |
895 | 0 | if (image->depth == 8) { |
896 | 0 | unormA = ptrA8[i]; |
897 | 0 | } else { |
898 | 0 | unormA = AVIF_MIN(ptrA16[i], yuvMaxChannel); |
899 | 0 | } |
900 | 0 | const float A = unormA / ((float)state->yuv.maxChannel); |
901 | 0 | const float Ac = AVIF_CLAMP(A, 0.0f, 1.0f); |
902 | |
|
903 | 0 | if (alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY) { |
904 | 0 | if (rgbHasColor) { |
905 | 0 | if (Ac == 0.0f) { |
906 | 0 | Rc = 0.0f; |
907 | 0 | Gc = 0.0f; |
908 | 0 | Bc = 0.0f; |
909 | 0 | } else if (Ac < 1.0f) { |
910 | 0 | Rc *= Ac; |
911 | 0 | Gc *= Ac; |
912 | 0 | Bc *= Ac; |
913 | 0 | } |
914 | 0 | } else { |
915 | 0 | if (Ac == 0.0f) { |
916 | 0 | grayc = 0.0f; |
917 | 0 | } else if (Ac < 1.0f) { |
918 | 0 | grayc *= Ac; |
919 | 0 | } |
920 | 0 | } |
921 | 0 | } else { |
922 | | // alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_UNMULTIPLY |
923 | 0 | if (rgbHasColor) { |
924 | 0 | if (Ac == 0.0f) { |
925 | 0 | Rc = 0.0f; |
926 | 0 | Gc = 0.0f; |
927 | 0 | Bc = 0.0f; |
928 | 0 | } else if (Ac < 1.0f) { |
929 | 0 | Rc /= Ac; |
930 | 0 | Gc /= Ac; |
931 | 0 | Bc /= Ac; |
932 | 0 | Rc = AVIF_MIN(Rc, 1.0f); |
933 | 0 | Gc = AVIF_MIN(Gc, 1.0f); |
934 | 0 | Bc = AVIF_MIN(Bc, 1.0f); |
935 | 0 | } |
936 | 0 | } else { |
937 | 0 | if (Ac == 0.0f) { |
938 | 0 | grayc = 0.0f; |
939 | 0 | } else if (Ac < 1.0f) { |
940 | 0 | grayc /= Ac; |
941 | 0 | grayc = AVIF_MIN(grayc, 1.0f); |
942 | 0 | } |
943 | 0 | } |
944 | 0 | } |
945 | 0 | } |
946 | | |
947 | 433M | if (rgbHasColor) { |
948 | 433M | if (rgb->depth == 8) { |
949 | 208M | avifStoreRGB8Pixel(rgb->format, |
950 | 208M | (uint8_t)(0.5f + (Rc * rgbMaxChannelF)), |
951 | 208M | (uint8_t)(0.5f + (Gc * rgbMaxChannelF)), |
952 | 208M | (uint8_t)(0.5f + (Bc * rgbMaxChannelF)), |
953 | 208M | ptrR, |
954 | 208M | ptrG, |
955 | 208M | ptrB); |
956 | 225M | } else { |
957 | 225M | *((uint16_t *)ptrR) = (uint16_t)(0.5f + (Rc * rgbMaxChannelF)); |
958 | 225M | *((uint16_t *)ptrG) = (uint16_t)(0.5f + (Gc * rgbMaxChannelF)); |
959 | 225M | *((uint16_t *)ptrB) = (uint16_t)(0.5f + (Bc * rgbMaxChannelF)); |
960 | 225M | } |
961 | 433M | ptrR += rgbPixelBytes; |
962 | 433M | ptrG += rgbPixelBytes; |
963 | 433M | ptrB += rgbPixelBytes; |
964 | 18.4E | } else { |
965 | 18.4E | if (rgb->depth == 8) { |
966 | 0 | *ptrGray = (uint8_t)(0.5f + (grayc * rgbMaxChannelF)); |
967 | 18.4E | } else { |
968 | 18.4E | *((uint16_t *)ptrGray) = (uint16_t)(0.5f + (grayc * rgbMaxChannelF)); |
969 | 18.4E | } |
970 | 18.4E | ptrGray += rgbPixelBytes; |
971 | 18.4E | } |
972 | 433M | } |
973 | 3.20M | } |
974 | 2.13k | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV); |
975 | 2.13k | return AVIF_RESULT_OK; |
976 | 2.13k | } |
977 | | |
978 | | static avifResult avifImageYUV16ToRGB16Color(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
979 | 1.43k | { |
980 | 1.43k | const float kr = state->yuv.kr; |
981 | 1.43k | const float kg = state->yuv.kg; |
982 | 1.43k | const float kb = state->yuv.kb; |
983 | 1.43k | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
984 | 1.43k | float * unormFloatTableY = NULL; |
985 | 1.43k | float * unormFloatTableUV = NULL; |
986 | 1.43k | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
987 | | |
988 | 1.43k | const uint16_t yuvMaxChannel = (uint16_t)state->yuv.maxChannel; |
989 | 1.43k | const float rgbMaxChannelF = state->rgb.maxChannelF; |
990 | 359k | for (size_t j = 0; j < image->height; ++j) { |
991 | 358k | const size_t uvJ = j >> state->yuv.formatInfo.chromaShiftY; |
992 | 358k | const uint16_t * const ptrY = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
993 | 358k | const uint16_t * const ptrU = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_U][(uvJ * image->yuvRowBytes[AVIF_CHAN_U])]; |
994 | 358k | const uint16_t * const ptrV = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_V][(uvJ * image->yuvRowBytes[AVIF_CHAN_V])]; |
995 | 358k | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
996 | 358k | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
997 | 358k | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
998 | | |
999 | 98.6M | for (size_t i = 0; i < image->width; ++i) { |
1000 | 98.3M | size_t uvI = i >> state->yuv.formatInfo.chromaShiftX; |
1001 | | |
1002 | | // clamp incoming data to protect against bad LUT lookups |
1003 | 98.3M | const uint16_t unormY = AVIF_MIN(ptrY[i], yuvMaxChannel); |
1004 | 98.3M | const uint16_t unormU = AVIF_MIN(ptrU[uvI], yuvMaxChannel); |
1005 | 98.3M | const uint16_t unormV = AVIF_MIN(ptrV[uvI], yuvMaxChannel); |
1006 | | |
1007 | | // Convert unorm to float |
1008 | 98.3M | const float Y = unormFloatTableY[unormY]; |
1009 | 98.3M | const float Cb = unormFloatTableUV[unormU]; |
1010 | 98.3M | const float Cr = unormFloatTableUV[unormV]; |
1011 | | |
1012 | 98.3M | const float R = Y + (2 * (1 - kr)) * Cr; |
1013 | 98.3M | const float B = Y + (2 * (1 - kb)) * Cb; |
1014 | 98.3M | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1015 | 98.3M | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1016 | 98.3M | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1017 | 98.3M | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1018 | | |
1019 | 98.3M | *((uint16_t *)ptrR) = (uint16_t)(0.5f + (Rc * rgbMaxChannelF)); |
1020 | 98.3M | *((uint16_t *)ptrG) = (uint16_t)(0.5f + (Gc * rgbMaxChannelF)); |
1021 | 98.3M | *((uint16_t *)ptrB) = (uint16_t)(0.5f + (Bc * rgbMaxChannelF)); |
1022 | | |
1023 | 98.3M | ptrR += rgbPixelBytes; |
1024 | 98.3M | ptrG += rgbPixelBytes; |
1025 | 98.3M | ptrB += rgbPixelBytes; |
1026 | 98.3M | } |
1027 | 358k | } |
1028 | 1.43k | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV); |
1029 | 1.43k | return AVIF_RESULT_OK; |
1030 | 1.43k | } |
1031 | | |
1032 | | static avifResult avifImageYUV16ToRGB16Mono(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1033 | 548 | { |
1034 | 548 | const float kr = state->yuv.kr; |
1035 | 548 | const float kg = state->yuv.kg; |
1036 | 548 | const float kb = state->yuv.kb; |
1037 | 548 | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1038 | 548 | float * unormFloatTableY = NULL; |
1039 | 548 | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, NULL, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1040 | | |
1041 | 548 | const uint16_t maxChannel = (uint16_t)state->yuv.maxChannel; |
1042 | 548 | const float maxChannelF = state->rgb.maxChannelF; |
1043 | 92.1k | for (size_t j = 0; j < image->height; ++j) { |
1044 | 91.6k | const uint16_t * const ptrY = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1045 | 91.6k | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1046 | 91.6k | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1047 | 91.6k | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1048 | | |
1049 | 165M | for (size_t i = 0; i < image->width; ++i) { |
1050 | | // clamp incoming data to protect against bad LUT lookups |
1051 | 165M | const uint16_t unormY = AVIF_MIN(ptrY[i], maxChannel); |
1052 | | |
1053 | | // Convert unorm to float |
1054 | 165M | const float Y = unormFloatTableY[unormY]; |
1055 | 165M | const float Cb = 0.0f; |
1056 | 165M | const float Cr = 0.0f; |
1057 | | |
1058 | 165M | const float R = Y + (2 * (1 - kr)) * Cr; |
1059 | 165M | const float B = Y + (2 * (1 - kb)) * Cb; |
1060 | 165M | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1061 | 165M | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1062 | 165M | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1063 | 165M | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1064 | | |
1065 | 165M | *((uint16_t *)ptrR) = (uint16_t)(0.5f + (Rc * maxChannelF)); |
1066 | 165M | *((uint16_t *)ptrG) = (uint16_t)(0.5f + (Gc * maxChannelF)); |
1067 | 165M | *((uint16_t *)ptrB) = (uint16_t)(0.5f + (Bc * maxChannelF)); |
1068 | | |
1069 | 165M | ptrR += rgbPixelBytes; |
1070 | 165M | ptrG += rgbPixelBytes; |
1071 | 165M | ptrB += rgbPixelBytes; |
1072 | 165M | } |
1073 | 91.6k | } |
1074 | 548 | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, NULL); |
1075 | 548 | return AVIF_RESULT_OK; |
1076 | 548 | } |
1077 | | |
1078 | | static avifResult avifImageYUV16ToRGB8Color(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1079 | 0 | { |
1080 | 0 | const float kr = state->yuv.kr; |
1081 | 0 | const float kg = state->yuv.kg; |
1082 | 0 | const float kb = state->yuv.kb; |
1083 | 0 | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1084 | 0 | float * unormFloatTableY = NULL; |
1085 | 0 | float * unormFloatTableUV = NULL; |
1086 | 0 | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1087 | | |
1088 | 0 | const uint16_t yuvMaxChannel = (uint16_t)state->yuv.maxChannel; |
1089 | 0 | const float rgbMaxChannelF = state->rgb.maxChannelF; |
1090 | 0 | for (size_t j = 0; j < image->height; ++j) { |
1091 | 0 | const size_t uvJ = j >> state->yuv.formatInfo.chromaShiftY; |
1092 | 0 | const uint16_t * const ptrY = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1093 | 0 | const uint16_t * const ptrU = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_U][(uvJ * image->yuvRowBytes[AVIF_CHAN_U])]; |
1094 | 0 | const uint16_t * const ptrV = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_V][(uvJ * image->yuvRowBytes[AVIF_CHAN_V])]; |
1095 | 0 | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1096 | 0 | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1097 | 0 | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1098 | |
|
1099 | 0 | for (size_t i = 0; i < image->width; ++i) { |
1100 | 0 | size_t uvI = i >> state->yuv.formatInfo.chromaShiftX; |
1101 | | |
1102 | | // clamp incoming data to protect against bad LUT lookups |
1103 | 0 | const uint16_t unormY = AVIF_MIN(ptrY[i], yuvMaxChannel); |
1104 | 0 | const uint16_t unormU = AVIF_MIN(ptrU[uvI], yuvMaxChannel); |
1105 | 0 | const uint16_t unormV = AVIF_MIN(ptrV[uvI], yuvMaxChannel); |
1106 | | |
1107 | | // Convert unorm to float |
1108 | 0 | const float Y = unormFloatTableY[unormY]; |
1109 | 0 | const float Cb = unormFloatTableUV[unormU]; |
1110 | 0 | const float Cr = unormFloatTableUV[unormV]; |
1111 | |
|
1112 | 0 | const float R = Y + (2 * (1 - kr)) * Cr; |
1113 | 0 | const float B = Y + (2 * (1 - kb)) * Cb; |
1114 | 0 | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1115 | 0 | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1116 | 0 | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1117 | 0 | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1118 | |
|
1119 | 0 | avifStoreRGB8Pixel(rgb->format, |
1120 | 0 | (uint8_t)(0.5f + (Rc * rgbMaxChannelF)), |
1121 | 0 | (uint8_t)(0.5f + (Gc * rgbMaxChannelF)), |
1122 | 0 | (uint8_t)(0.5f + (Bc * rgbMaxChannelF)), |
1123 | 0 | ptrR, |
1124 | 0 | ptrG, |
1125 | 0 | ptrB); |
1126 | |
|
1127 | 0 | ptrR += rgbPixelBytes; |
1128 | 0 | ptrG += rgbPixelBytes; |
1129 | 0 | ptrB += rgbPixelBytes; |
1130 | 0 | } |
1131 | 0 | } |
1132 | 0 | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV); |
1133 | 0 | return AVIF_RESULT_OK; |
1134 | 0 | } |
1135 | | |
1136 | | static avifResult avifImageYUV16ToRGB8Mono(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1137 | 0 | { |
1138 | 0 | const float kr = state->yuv.kr; |
1139 | 0 | const float kg = state->yuv.kg; |
1140 | 0 | const float kb = state->yuv.kb; |
1141 | 0 | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1142 | 0 | float * unormFloatTableY = NULL; |
1143 | 0 | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, NULL, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1144 | | |
1145 | 0 | const uint16_t yuvMaxChannel = (uint16_t)state->yuv.maxChannel; |
1146 | 0 | const float rgbMaxChannelF = state->rgb.maxChannelF; |
1147 | 0 | for (size_t j = 0; j < image->height; ++j) { |
1148 | 0 | const uint16_t * const ptrY = (uint16_t *)&image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1149 | 0 | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1150 | 0 | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1151 | 0 | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1152 | |
|
1153 | 0 | for (size_t i = 0; i < image->width; ++i) { |
1154 | | // clamp incoming data to protect against bad LUT lookups |
1155 | 0 | const uint16_t unormY = AVIF_MIN(ptrY[i], yuvMaxChannel); |
1156 | | |
1157 | | // Convert unorm to float |
1158 | 0 | const float Y = unormFloatTableY[unormY]; |
1159 | 0 | const float Cb = 0.0f; |
1160 | 0 | const float Cr = 0.0f; |
1161 | |
|
1162 | 0 | const float R = Y + (2 * (1 - kr)) * Cr; |
1163 | 0 | const float B = Y + (2 * (1 - kb)) * Cb; |
1164 | 0 | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1165 | 0 | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1166 | 0 | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1167 | 0 | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1168 | |
|
1169 | 0 | avifStoreRGB8Pixel(rgb->format, |
1170 | 0 | (uint8_t)(0.5f + (Rc * rgbMaxChannelF)), |
1171 | 0 | (uint8_t)(0.5f + (Gc * rgbMaxChannelF)), |
1172 | 0 | (uint8_t)(0.5f + (Bc * rgbMaxChannelF)), |
1173 | 0 | ptrR, |
1174 | 0 | ptrG, |
1175 | 0 | ptrB); |
1176 | |
|
1177 | 0 | ptrR += rgbPixelBytes; |
1178 | 0 | ptrG += rgbPixelBytes; |
1179 | 0 | ptrB += rgbPixelBytes; |
1180 | 0 | } |
1181 | 0 | } |
1182 | 0 | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, NULL); |
1183 | 0 | return AVIF_RESULT_OK; |
1184 | 0 | } |
1185 | | |
1186 | | static avifResult avifImageYUV8ToRGB16Color(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1187 | 0 | { |
1188 | 0 | const float kr = state->yuv.kr; |
1189 | 0 | const float kg = state->yuv.kg; |
1190 | 0 | const float kb = state->yuv.kb; |
1191 | 0 | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1192 | 0 | float * unormFloatTableY = NULL; |
1193 | 0 | float * unormFloatTableUV = NULL; |
1194 | 0 | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1195 | | |
1196 | 0 | const float rgbMaxChannelF = state->rgb.maxChannelF; |
1197 | 0 | for (size_t j = 0; j < image->height; ++j) { |
1198 | 0 | const size_t uvJ = j >> state->yuv.formatInfo.chromaShiftY; |
1199 | 0 | const uint8_t * const ptrY = &image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1200 | 0 | const uint8_t * const ptrU = &image->yuvPlanes[AVIF_CHAN_U][(uvJ * image->yuvRowBytes[AVIF_CHAN_U])]; |
1201 | 0 | const uint8_t * const ptrV = &image->yuvPlanes[AVIF_CHAN_V][(uvJ * image->yuvRowBytes[AVIF_CHAN_V])]; |
1202 | 0 | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1203 | 0 | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1204 | 0 | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1205 | |
|
1206 | 0 | for (size_t i = 0; i < image->width; ++i) { |
1207 | 0 | size_t uvI = i >> state->yuv.formatInfo.chromaShiftX; |
1208 | | |
1209 | | // Convert unorm to float (no clamp necessary, the full uint8_t range is a legal lookup) |
1210 | 0 | const float Y = unormFloatTableY[ptrY[i]]; |
1211 | 0 | const float Cb = unormFloatTableUV[ptrU[uvI]]; |
1212 | 0 | const float Cr = unormFloatTableUV[ptrV[uvI]]; |
1213 | |
|
1214 | 0 | const float R = Y + (2 * (1 - kr)) * Cr; |
1215 | 0 | const float B = Y + (2 * (1 - kb)) * Cb; |
1216 | 0 | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1217 | 0 | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1218 | 0 | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1219 | 0 | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1220 | |
|
1221 | 0 | *((uint16_t *)ptrR) = (uint16_t)(0.5f + (Rc * rgbMaxChannelF)); |
1222 | 0 | *((uint16_t *)ptrG) = (uint16_t)(0.5f + (Gc * rgbMaxChannelF)); |
1223 | 0 | *((uint16_t *)ptrB) = (uint16_t)(0.5f + (Bc * rgbMaxChannelF)); |
1224 | |
|
1225 | 0 | ptrR += rgbPixelBytes; |
1226 | 0 | ptrG += rgbPixelBytes; |
1227 | 0 | ptrB += rgbPixelBytes; |
1228 | 0 | } |
1229 | 0 | } |
1230 | 0 | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV); |
1231 | 0 | return AVIF_RESULT_OK; |
1232 | 0 | } |
1233 | | |
1234 | | static avifResult avifImageYUV8ToRGB16Mono(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1235 | 0 | { |
1236 | 0 | const float kr = state->yuv.kr; |
1237 | 0 | const float kg = state->yuv.kg; |
1238 | 0 | const float kb = state->yuv.kb; |
1239 | 0 | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1240 | 0 | float * unormFloatTableY = NULL; |
1241 | 0 | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, NULL, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1242 | | |
1243 | 0 | const float rgbMaxChannelF = state->rgb.maxChannelF; |
1244 | 0 | for (size_t j = 0; j < image->height; ++j) { |
1245 | 0 | const uint8_t * const ptrY = &image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1246 | 0 | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1247 | 0 | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1248 | 0 | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1249 | |
|
1250 | 0 | for (size_t i = 0; i < image->width; ++i) { |
1251 | | // Convert unorm to float (no clamp necessary, the full uint8_t range is a legal lookup) |
1252 | 0 | const float Y = unormFloatTableY[ptrY[i]]; |
1253 | 0 | const float Cb = 0.0f; |
1254 | 0 | const float Cr = 0.0f; |
1255 | |
|
1256 | 0 | const float R = Y + (2 * (1 - kr)) * Cr; |
1257 | 0 | const float B = Y + (2 * (1 - kb)) * Cb; |
1258 | 0 | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1259 | 0 | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1260 | 0 | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1261 | 0 | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1262 | |
|
1263 | 0 | *((uint16_t *)ptrR) = (uint16_t)(0.5f + (Rc * rgbMaxChannelF)); |
1264 | 0 | *((uint16_t *)ptrG) = (uint16_t)(0.5f + (Gc * rgbMaxChannelF)); |
1265 | 0 | *((uint16_t *)ptrB) = (uint16_t)(0.5f + (Bc * rgbMaxChannelF)); |
1266 | |
|
1267 | 0 | ptrR += rgbPixelBytes; |
1268 | 0 | ptrG += rgbPixelBytes; |
1269 | 0 | ptrB += rgbPixelBytes; |
1270 | 0 | } |
1271 | 0 | } |
1272 | 0 | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, NULL); |
1273 | 0 | return AVIF_RESULT_OK; |
1274 | 0 | } |
1275 | | |
1276 | | static avifResult avifImageIdentity8ToRGB8ColorFullRange(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1277 | 6.50k | { |
1278 | 6.50k | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1279 | 740k | for (size_t j = 0; j < image->height; ++j) { |
1280 | 733k | const uint8_t * const ptrY = &image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1281 | 733k | const uint8_t * const ptrU = &image->yuvPlanes[AVIF_CHAN_U][(j * image->yuvRowBytes[AVIF_CHAN_U])]; |
1282 | 733k | const uint8_t * const ptrV = &image->yuvPlanes[AVIF_CHAN_V][(j * image->yuvRowBytes[AVIF_CHAN_V])]; |
1283 | 733k | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1284 | 733k | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1285 | 733k | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1286 | | |
1287 | | // This is intentionally a per-row conditional instead of a per-pixel |
1288 | | // conditional. This makes the "else" path (much more common than the |
1289 | | // "if" path) much faster than having a per-pixel branch. |
1290 | 733k | if (rgb->format == AVIF_RGB_FORMAT_RGB_565) { |
1291 | 0 | for (size_t i = 0; i < image->width; ++i) { |
1292 | 0 | *(uint16_t *)ptrR = RGB565(ptrV[i], ptrY[i], ptrU[i]); |
1293 | 0 | ptrR += rgbPixelBytes; |
1294 | 0 | } |
1295 | 733k | } else { |
1296 | 109M | for (size_t i = 0; i < image->width; ++i) { |
1297 | 108M | *ptrR = ptrV[i]; |
1298 | 108M | *ptrG = ptrY[i]; |
1299 | 108M | *ptrB = ptrU[i]; |
1300 | 108M | ptrR += rgbPixelBytes; |
1301 | 108M | ptrG += rgbPixelBytes; |
1302 | 108M | ptrB += rgbPixelBytes; |
1303 | 108M | } |
1304 | 733k | } |
1305 | 733k | } |
1306 | 6.50k | return AVIF_RESULT_OK; |
1307 | 6.50k | } |
1308 | | |
1309 | | static avifResult avifImageYUV8ToRGB8Color(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1310 | 1.46k | { |
1311 | 1.46k | const float kr = state->yuv.kr; |
1312 | 1.46k | const float kg = state->yuv.kg; |
1313 | 1.46k | const float kb = state->yuv.kb; |
1314 | 1.46k | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1315 | 1.46k | float * unormFloatTableY = NULL; |
1316 | 1.46k | float * unormFloatTableUV = NULL; |
1317 | 1.46k | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1318 | | |
1319 | 1.46k | const float rgbMaxChannelF = state->rgb.maxChannelF; |
1320 | 744k | for (size_t j = 0; j < image->height; ++j) { |
1321 | 743k | const size_t uvJ = j >> state->yuv.formatInfo.chromaShiftY; |
1322 | 743k | const uint8_t * const ptrY = &image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1323 | 743k | const uint8_t * const ptrU = &image->yuvPlanes[AVIF_CHAN_U][(uvJ * image->yuvRowBytes[AVIF_CHAN_U])]; |
1324 | 743k | const uint8_t * const ptrV = &image->yuvPlanes[AVIF_CHAN_V][(uvJ * image->yuvRowBytes[AVIF_CHAN_V])]; |
1325 | 743k | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1326 | 743k | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1327 | 743k | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1328 | | |
1329 | 149M | for (size_t i = 0; i < image->width; ++i) { |
1330 | 148M | size_t uvI = i >> state->yuv.formatInfo.chromaShiftX; |
1331 | | |
1332 | | // Convert unorm to float (no clamp necessary, the full uint8_t range is a legal lookup) |
1333 | 148M | const float Y = unormFloatTableY[ptrY[i]]; |
1334 | 148M | const float Cb = unormFloatTableUV[ptrU[uvI]]; |
1335 | 148M | const float Cr = unormFloatTableUV[ptrV[uvI]]; |
1336 | | |
1337 | 148M | const float R = Y + (2 * (1 - kr)) * Cr; |
1338 | 148M | const float B = Y + (2 * (1 - kb)) * Cb; |
1339 | 148M | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1340 | 148M | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1341 | 148M | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1342 | 148M | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1343 | | |
1344 | 148M | avifStoreRGB8Pixel(rgb->format, |
1345 | 148M | (uint8_t)(0.5f + (Rc * rgbMaxChannelF)), |
1346 | 148M | (uint8_t)(0.5f + (Gc * rgbMaxChannelF)), |
1347 | 148M | (uint8_t)(0.5f + (Bc * rgbMaxChannelF)), |
1348 | 148M | ptrR, |
1349 | 148M | ptrG, |
1350 | 148M | ptrB); |
1351 | | |
1352 | 148M | ptrR += rgbPixelBytes; |
1353 | 148M | ptrG += rgbPixelBytes; |
1354 | 148M | ptrB += rgbPixelBytes; |
1355 | 148M | } |
1356 | 743k | } |
1357 | 1.46k | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, &unormFloatTableUV); |
1358 | 1.46k | return AVIF_RESULT_OK; |
1359 | 1.46k | } |
1360 | | |
1361 | | static avifResult avifImageYUV8ToRGB8Mono(const avifImage * image, avifRGBImage * rgb, avifReformatState * state) |
1362 | 729 | { |
1363 | 729 | const float kr = state->yuv.kr; |
1364 | 729 | const float kg = state->yuv.kg; |
1365 | 729 | const float kb = state->yuv.kb; |
1366 | 729 | const uint32_t rgbPixelBytes = state->rgb.pixelBytes; |
1367 | 729 | float * unormFloatTableY = NULL; |
1368 | 729 | AVIF_CHECKERR(avifCreateYUVToRGBLookUpTables(&unormFloatTableY, NULL, image->depth, state), AVIF_RESULT_OUT_OF_MEMORY); |
1369 | | |
1370 | 729 | const float rgbMaxChannelF = state->rgb.maxChannelF; |
1371 | 255k | for (size_t j = 0; j < image->height; ++j) { |
1372 | 254k | const uint8_t * const ptrY = &image->yuvPlanes[AVIF_CHAN_Y][(j * image->yuvRowBytes[AVIF_CHAN_Y])]; |
1373 | 254k | uint8_t * ptrR = &rgb->pixels[state->rgb.offsetBytesR + (j * rgb->rowBytes)]; |
1374 | 254k | uint8_t * ptrG = &rgb->pixels[state->rgb.offsetBytesG + (j * rgb->rowBytes)]; |
1375 | 254k | uint8_t * ptrB = &rgb->pixels[state->rgb.offsetBytesB + (j * rgb->rowBytes)]; |
1376 | | |
1377 | 461M | for (size_t i = 0; i < image->width; ++i) { |
1378 | | // Convert unorm to float (no clamp necessary, the full uint8_t range is a legal lookup) |
1379 | 461M | const float Y = unormFloatTableY[ptrY[i]]; |
1380 | 461M | const float Cb = 0.0f; |
1381 | 461M | const float Cr = 0.0f; |
1382 | | |
1383 | 461M | const float R = Y + (2 * (1 - kr)) * Cr; |
1384 | 461M | const float B = Y + (2 * (1 - kb)) * Cb; |
1385 | 461M | const float G = Y - ((2 * ((kr * (1 - kr) * Cr) + (kb * (1 - kb) * Cb))) / kg); |
1386 | 461M | const float Rc = AVIF_CLAMP(R, 0.0f, 1.0f); |
1387 | 461M | const float Gc = AVIF_CLAMP(G, 0.0f, 1.0f); |
1388 | 461M | const float Bc = AVIF_CLAMP(B, 0.0f, 1.0f); |
1389 | | |
1390 | 461M | avifStoreRGB8Pixel(rgb->format, |
1391 | 461M | (uint8_t)(0.5f + (Rc * rgbMaxChannelF)), |
1392 | 461M | (uint8_t)(0.5f + (Gc * rgbMaxChannelF)), |
1393 | 461M | (uint8_t)(0.5f + (Bc * rgbMaxChannelF)), |
1394 | 461M | ptrR, |
1395 | 461M | ptrG, |
1396 | 461M | ptrB); |
1397 | | |
1398 | 461M | ptrR += rgbPixelBytes; |
1399 | 461M | ptrG += rgbPixelBytes; |
1400 | 461M | ptrB += rgbPixelBytes; |
1401 | 461M | } |
1402 | 254k | } |
1403 | 729 | avifFreeYUVToRGBLookUpTables(&unormFloatTableY, NULL); |
1404 | 729 | return AVIF_RESULT_OK; |
1405 | 729 | } |
1406 | | |
1407 | | // This constant comes from libyuv. For details, see here: |
1408 | | // https://chromium.googlesource.com/libyuv/libyuv/+/2f87e9a7/source/row_common.cc#3537 |
1409 | 0 | #define F16_MULTIPLIER 1.9259299444e-34f |
1410 | | |
1411 | | typedef union avifF16 |
1412 | | { |
1413 | | float f; |
1414 | | uint32_t u32; |
1415 | | } avifF16; |
1416 | | |
1417 | | static avifResult avifRGBImageToF16(avifRGBImage * rgb) |
1418 | 0 | { |
1419 | 0 | avifResult libyuvResult = AVIF_RESULT_NOT_IMPLEMENTED; |
1420 | 0 | if (!rgb->avoidLibYUV) { |
1421 | 0 | libyuvResult = avifRGBImageToF16LibYUV(rgb); |
1422 | 0 | } |
1423 | 0 | if (libyuvResult != AVIF_RESULT_NOT_IMPLEMENTED) { |
1424 | 0 | return libyuvResult; |
1425 | 0 | } |
1426 | 0 | const size_t channelCount = avifRGBFormatChannelCount(rgb->format); |
1427 | 0 | const float scale = 1.0f / ((1 << rgb->depth) - 1); |
1428 | 0 | const float multiplier = F16_MULTIPLIER * scale; |
1429 | 0 | uint16_t * pixelRowBase = (uint16_t *)rgb->pixels; |
1430 | 0 | const uint32_t stride = rgb->rowBytes >> 1; |
1431 | 0 | for (size_t j = 0; j < rgb->height; ++j) { |
1432 | 0 | uint16_t * pixel = pixelRowBase; |
1433 | 0 | for (size_t i = 0; i < rgb->width * channelCount; ++i, ++pixel) { |
1434 | 0 | avifF16 f16; |
1435 | 0 | f16.f = *pixel * multiplier; |
1436 | 0 | *pixel = (uint16_t)(f16.u32 >> 13); |
1437 | 0 | } |
1438 | 0 | pixelRowBase += stride; |
1439 | 0 | } |
1440 | 0 | return AVIF_RESULT_OK; |
1441 | 0 | } |
1442 | | |
1443 | | static avifResult avifImageYUVToRGBImpl(const avifImage * image, avifRGBImage * rgb, avifReformatState * state, avifAlphaMultiplyMode alphaMultiplyMode) |
1444 | 12.8k | { |
1445 | 12.8k | avifBool convertedWithLibYUV = AVIF_FALSE; |
1446 | | // Reformat alpha, if user asks for it, or (un)multiply processing needs it. |
1447 | 12.8k | avifBool reformatAlpha = avifRGBFormatHasAlpha(rgb->format) && |
1448 | 12.8k | (!rgb->ignoreAlpha || (alphaMultiplyMode != AVIF_ALPHA_MULTIPLY_MODE_NO_OP)); |
1449 | | // This value is used only when reformatAlpha is true. |
1450 | 12.8k | avifBool alphaReformattedWithLibYUV = AVIF_FALSE; |
1451 | 12.8k | if (!rgb->avoidLibYUV && ((alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_NO_OP) || avifRGBFormatHasAlpha(rgb->format))) { |
1452 | 12.8k | avifResult libyuvResult = avifImageYUVToRGBLibYUV(image, rgb, reformatAlpha, &alphaReformattedWithLibYUV); |
1453 | 12.8k | if (libyuvResult == AVIF_RESULT_OK) { |
1454 | 0 | convertedWithLibYUV = AVIF_TRUE; |
1455 | 12.8k | } else { |
1456 | 12.8k | if (libyuvResult != AVIF_RESULT_NOT_IMPLEMENTED) { |
1457 | 0 | return libyuvResult; |
1458 | 0 | } |
1459 | 12.8k | } |
1460 | 12.8k | } |
1461 | | |
1462 | 12.8k | if (reformatAlpha && !alphaReformattedWithLibYUV) { |
1463 | 12.7k | avifAlphaParams params; |
1464 | | |
1465 | 12.7k | params.width = rgb->width; |
1466 | 12.7k | params.height = rgb->height; |
1467 | 12.7k | params.dstDepth = rgb->depth; |
1468 | 12.7k | params.dstPlane = rgb->pixels; |
1469 | 12.7k | params.dstRowBytes = rgb->rowBytes; |
1470 | 12.7k | params.dstOffsetBytes = state->rgb.offsetBytesA; |
1471 | 12.7k | params.dstPixelBytes = state->rgb.pixelBytes; |
1472 | | |
1473 | 12.7k | if (image->alphaPlane && image->alphaRowBytes) { |
1474 | 251 | params.srcDepth = image->depth; |
1475 | 251 | params.srcPlane = image->alphaPlane; |
1476 | 251 | params.srcRowBytes = image->alphaRowBytes; |
1477 | 251 | params.srcOffsetBytes = 0; |
1478 | 251 | params.srcPixelBytes = state->yuv.channelBytes; |
1479 | | |
1480 | 251 | avifReformatAlpha(¶ms); |
1481 | 12.5k | } else { |
1482 | 12.5k | avifFillAlpha(¶ms); |
1483 | 12.5k | } |
1484 | 12.7k | } |
1485 | | |
1486 | 12.8k | if (!convertedWithLibYUV) { |
1487 | | // libyuv is either unavailable or unable to perform the specific conversion required here. |
1488 | | // Look over the available built-in "fast" routines for YUV->RGB conversion and see if one |
1489 | | // fits the current combination, or as a last resort, call avifImageYUVAnyToRGBAnySlow(), |
1490 | | // which handles every possibly YUV->RGB combination, but very slowly (in comparison). |
1491 | | |
1492 | 12.8k | avifResult convertResult = AVIF_RESULT_NOT_IMPLEMENTED; |
1493 | | |
1494 | 12.8k | const avifBool hasColor = |
1495 | 12.8k | (image->yuvRowBytes[AVIF_CHAN_U] && image->yuvRowBytes[AVIF_CHAN_V] && (image->yuvFormat != AVIF_PIXEL_FORMAT_YUV400)); |
1496 | | |
1497 | 12.8k | if (!avifRGBFormatIsGray(rgb->format) && |
1498 | 12.8k | (!hasColor || (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV444) || |
1499 | 1.00k | ((rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_FASTEST) || (rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_NEAREST))) && |
1500 | 11.8k | (alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_NO_OP || avifRGBFormatHasAlpha(rgb->format))) { |
1501 | | // Explanations on the above conditional: |
1502 | | // * None of these fast paths currently support bilinear upsampling, so avoid all of them |
1503 | | // unless the YUV data isn't subsampled or they explicitly requested AVIF_CHROMA_UPSAMPLING_NEAREST. |
1504 | | // * None of these fast paths currently handle alpha (un)multiply, so avoid all of them |
1505 | | // if we can't do alpha (un)multiply as a separated post step (destination format doesn't have alpha). |
1506 | | |
1507 | 11.8k | if (state->yuv.mode == AVIF_REFORMAT_MODE_IDENTITY) { |
1508 | 6.88k | if ((image->depth == 8) && (rgb->depth == 8) && (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV444) && |
1509 | 6.53k | (image->yuvRange == AVIF_RANGE_FULL)) { |
1510 | 6.50k | convertResult = avifImageIdentity8ToRGB8ColorFullRange(image, rgb, state); |
1511 | 6.50k | } |
1512 | | |
1513 | | // TODO: Add more fast paths for identity |
1514 | 6.88k | } else if (state->yuv.mode == AVIF_REFORMAT_MODE_YUV_COEFFICIENTS) { |
1515 | 4.16k | if (image->depth > 8) { |
1516 | | // yuv:u16 |
1517 | | |
1518 | 1.97k | if (rgb->depth > 8) { |
1519 | | // yuv:u16, rgb:u16 |
1520 | | |
1521 | 1.97k | if (hasColor) { |
1522 | 1.43k | convertResult = avifImageYUV16ToRGB16Color(image, rgb, state); |
1523 | 1.43k | } else { |
1524 | 548 | convertResult = avifImageYUV16ToRGB16Mono(image, rgb, state); |
1525 | 548 | } |
1526 | 1.97k | } else { |
1527 | | // yuv:u16, rgb:u8 |
1528 | |
|
1529 | 0 | if (hasColor) { |
1530 | 0 | convertResult = avifImageYUV16ToRGB8Color(image, rgb, state); |
1531 | 0 | } else { |
1532 | 0 | convertResult = avifImageYUV16ToRGB8Mono(image, rgb, state); |
1533 | 0 | } |
1534 | 0 | } |
1535 | 2.19k | } else { |
1536 | | // yuv:u8 |
1537 | | |
1538 | 2.19k | if (rgb->depth > 8) { |
1539 | | // yuv:u8, rgb:u16 |
1540 | |
|
1541 | 0 | if (hasColor) { |
1542 | 0 | convertResult = avifImageYUV8ToRGB16Color(image, rgb, state); |
1543 | 0 | } else { |
1544 | 0 | convertResult = avifImageYUV8ToRGB16Mono(image, rgb, state); |
1545 | 0 | } |
1546 | 2.19k | } else { |
1547 | | // yuv:u8, rgb:u8 |
1548 | | |
1549 | 2.19k | if (hasColor) { |
1550 | 1.46k | convertResult = avifImageYUV8ToRGB8Color(image, rgb, state); |
1551 | 1.46k | } else { |
1552 | 729 | convertResult = avifImageYUV8ToRGB8Mono(image, rgb, state); |
1553 | 729 | } |
1554 | 2.19k | } |
1555 | 2.19k | } |
1556 | 4.16k | } |
1557 | 11.8k | } |
1558 | | |
1559 | 12.8k | if (convertResult == AVIF_RESULT_NOT_IMPLEMENTED) { |
1560 | | // If we get here, there is no fast path for this combination. Time to be slow! |
1561 | 2.13k | convertResult = avifImageYUVAnyToRGBAnySlow(image, rgb, state, alphaMultiplyMode); |
1562 | | |
1563 | | // The slow path also handles alpha (un)multiply, so forget the operation here. |
1564 | 2.13k | alphaMultiplyMode = AVIF_ALPHA_MULTIPLY_MODE_NO_OP; |
1565 | 2.13k | } |
1566 | | |
1567 | 12.8k | if (convertResult != AVIF_RESULT_OK) { |
1568 | 0 | return convertResult; |
1569 | 0 | } |
1570 | 12.8k | } |
1571 | | |
1572 | | // Process alpha premultiplication, if necessary |
1573 | 12.8k | if (alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY) { |
1574 | 0 | avifResult result = avifRGBImagePremultiplyAlpha(rgb); |
1575 | 0 | if (result != AVIF_RESULT_OK) { |
1576 | 0 | return result; |
1577 | 0 | } |
1578 | 12.8k | } else if (alphaMultiplyMode == AVIF_ALPHA_MULTIPLY_MODE_UNMULTIPLY) { |
1579 | 0 | avifResult result = avifRGBImageUnpremultiplyAlpha(rgb); |
1580 | 0 | if (result != AVIF_RESULT_OK) { |
1581 | 0 | return result; |
1582 | 0 | } |
1583 | 0 | } |
1584 | | |
1585 | | // Convert pixels to half floats (F16), if necessary. |
1586 | 12.8k | if (rgb->isFloat) { |
1587 | 0 | return avifRGBImageToF16(rgb); |
1588 | 0 | } |
1589 | | |
1590 | 12.8k | return AVIF_RESULT_OK; |
1591 | 12.8k | } |
1592 | | |
1593 | | typedef struct |
1594 | | { |
1595 | | #if defined(_WIN32) |
1596 | | HANDLE thread; |
1597 | | #else |
1598 | | pthread_t thread; |
1599 | | #endif |
1600 | | avifImage image; |
1601 | | avifRGBImage rgb; |
1602 | | avifReformatState * state; |
1603 | | avifAlphaMultiplyMode alphaMultiplyMode; |
1604 | | avifResult result; |
1605 | | avifBool threadCreated; |
1606 | | } YUVToRGBThreadData; |
1607 | | |
1608 | | #if defined(_WIN32) |
1609 | | static unsigned int __stdcall avifImageYUVToRGBThreadWorker(void * arg) |
1610 | | #else |
1611 | | static void * avifImageYUVToRGBThreadWorker(void * arg) |
1612 | | #endif |
1613 | 12.0k | { |
1614 | 12.0k | YUVToRGBThreadData * data = (YUVToRGBThreadData *)arg; |
1615 | 12.0k | data->result = avifImageYUVToRGBImpl(&data->image, &data->rgb, data->state, data->alphaMultiplyMode); |
1616 | | #if defined(_WIN32) |
1617 | | return 0; |
1618 | | #else |
1619 | 12.0k | return NULL; |
1620 | 12.0k | #endif |
1621 | 12.0k | } |
1622 | | |
1623 | | static avifBool avifCreateYUVToRGBThread(YUVToRGBThreadData * tdata) |
1624 | 10.5k | { |
1625 | | #if defined(_WIN32) |
1626 | | tdata->thread = (HANDLE)_beginthreadex(/*security=*/NULL, |
1627 | | /*stack_size=*/0, |
1628 | | &avifImageYUVToRGBThreadWorker, |
1629 | | tdata, |
1630 | | /*initflag=*/0, |
1631 | | /*thrdaddr=*/NULL); |
1632 | | return tdata->thread != NULL; |
1633 | | #else |
1634 | 10.5k | return pthread_create(&tdata->thread, NULL, &avifImageYUVToRGBThreadWorker, tdata) == 0; |
1635 | 10.5k | #endif |
1636 | 10.5k | } |
1637 | | |
1638 | | static avifBool avifJoinYUVToRGBThread(YUVToRGBThreadData * tdata) |
1639 | 10.5k | { |
1640 | | #if defined(_WIN32) |
1641 | | return WaitForSingleObject(tdata->thread, INFINITE) == WAIT_OBJECT_0 && CloseHandle(tdata->thread) != 0; |
1642 | | #else |
1643 | 10.5k | return pthread_join(tdata->thread, NULL) == 0; |
1644 | 10.5k | #endif |
1645 | 10.5k | } |
1646 | | |
1647 | | avifResult avifImageYUVToRGB(const avifImage * image, avifRGBImage * rgb) |
1648 | 2.38k | { |
1649 | | // It is okay for rgb->maxThreads to be equal to zero in order to allow clients to zero initialize the avifRGBImage struct |
1650 | | // with memset. |
1651 | 2.38k | if (!image->yuvPlanes[AVIF_CHAN_Y] || rgb->maxThreads < 0) { |
1652 | 0 | return AVIF_RESULT_REFORMAT_FAILED; |
1653 | 0 | } |
1654 | | |
1655 | 2.38k | avifReformatState state; |
1656 | 2.38k | if (!avifPrepareReformatState(image, rgb, &state)) { |
1657 | 159 | return AVIF_RESULT_REFORMAT_FAILED; |
1658 | 159 | } |
1659 | | |
1660 | 2.22k | avifAlphaMultiplyMode alphaMultiplyMode = AVIF_ALPHA_MULTIPLY_MODE_NO_OP; |
1661 | 2.22k | if (image->alphaPlane) { |
1662 | 48 | if (!avifRGBFormatHasAlpha(rgb->format) || rgb->ignoreAlpha) { |
1663 | | // if we are converting some image with alpha into a format without alpha, we should do 'premultiply alpha' before |
1664 | | // discarding alpha plane. This has the same effect of rendering this image on a black background, which makes sense. |
1665 | 0 | if (!image->alphaPremultiplied) { |
1666 | 0 | alphaMultiplyMode = AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY; |
1667 | 0 | } |
1668 | 48 | } else { |
1669 | 48 | if (!image->alphaPremultiplied && rgb->alphaPremultiplied) { |
1670 | 0 | alphaMultiplyMode = AVIF_ALPHA_MULTIPLY_MODE_MULTIPLY; |
1671 | 48 | } else if (image->alphaPremultiplied && !rgb->alphaPremultiplied) { |
1672 | 0 | alphaMultiplyMode = AVIF_ALPHA_MULTIPLY_MODE_UNMULTIPLY; |
1673 | 0 | } |
1674 | 48 | } |
1675 | 48 | } |
1676 | | |
1677 | | // In practice, we rarely need more than 8 threads for YUV to RGB conversion. |
1678 | 2.22k | uint32_t jobs = AVIF_CLAMP(rgb->maxThreads, 1, 8); |
1679 | | |
1680 | | // When yuv format is 420 and chromaUpsampling could be BILINEAR, there is a dependency across the horizontal borders of each |
1681 | | // job. So we disallow multithreading in that case. |
1682 | 2.22k | if (image->yuvFormat == AVIF_PIXEL_FORMAT_YUV420 && (rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_AUTOMATIC || |
1683 | 0 | rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_BEST_QUALITY || |
1684 | 217 | rgb->chromaUpsampling == AVIF_CHROMA_UPSAMPLING_BILINEAR)) { |
1685 | 217 | jobs = 1; |
1686 | 217 | } |
1687 | | |
1688 | | // Each thread worker needs at least 2 Y rows (to account for potential U/V subsampling). |
1689 | 2.22k | if (jobs == 1 || (image->height / 2) < jobs) { |
1690 | 706 | return avifImageYUVToRGBImpl(image, rgb, &state, alphaMultiplyMode); |
1691 | 706 | } |
1692 | | |
1693 | 1.51k | const size_t byteCount = sizeof(YUVToRGBThreadData) * jobs; |
1694 | 1.51k | YUVToRGBThreadData * threadData = (YUVToRGBThreadData *)avifAlloc(byteCount); |
1695 | 1.51k | if (!threadData) { |
1696 | 0 | return AVIF_RESULT_OUT_OF_MEMORY; |
1697 | 0 | } |
1698 | 1.51k | memset(threadData, 0, byteCount); |
1699 | 1.51k | uint32_t rowsPerJob = image->height / jobs; |
1700 | 1.51k | if (rowsPerJob % 2) { |
1701 | 248 | ++rowsPerJob; |
1702 | 248 | jobs = (image->height + rowsPerJob - 1) / rowsPerJob; // ceil |
1703 | 248 | } |
1704 | 1.51k | const uint32_t rowsForLastJob = image->height - rowsPerJob * (jobs - 1); |
1705 | 1.51k | uint32_t startRow = 0; |
1706 | 1.51k | uint32_t i; |
1707 | 13.6k | for (i = 0; i < jobs; ++i, startRow += rowsPerJob) { |
1708 | 12.0k | YUVToRGBThreadData * tdata = &threadData[i]; |
1709 | 12.0k | const avifCropRect rect = { .x = 0, .y = startRow, .width = image->width, .height = (i == jobs - 1) ? rowsForLastJob : rowsPerJob }; |
1710 | 12.0k | if (avifImageSetViewRect(&tdata->image, image, &rect) != AVIF_RESULT_OK) { |
1711 | 0 | tdata->result = AVIF_RESULT_REFORMAT_FAILED; |
1712 | 0 | break; |
1713 | 0 | } |
1714 | | |
1715 | 12.0k | tdata->rgb = *rgb; |
1716 | 12.0k | tdata->rgb.pixels += startRow * (size_t)rgb->rowBytes; |
1717 | 12.0k | tdata->rgb.height = tdata->image.height; |
1718 | | |
1719 | 12.0k | tdata->state = &state; |
1720 | 12.0k | tdata->alphaMultiplyMode = alphaMultiplyMode; |
1721 | | |
1722 | 12.0k | if (i > 0) { |
1723 | 10.5k | tdata->threadCreated = avifCreateYUVToRGBThread(tdata); |
1724 | 10.5k | if (!tdata->threadCreated) { |
1725 | 0 | tdata->result = AVIF_RESULT_REFORMAT_FAILED; |
1726 | 0 | break; |
1727 | 0 | } |
1728 | 10.5k | } |
1729 | 12.0k | } |
1730 | | // If above loop ran successfully, run the first job in the current thread. |
1731 | 1.51k | if (i == jobs) { |
1732 | 1.51k | avifImageYUVToRGBThreadWorker(&threadData[0]); |
1733 | 1.51k | } |
1734 | 1.51k | avifResult result = AVIF_RESULT_OK; |
1735 | 13.6k | for (i = 0; i < jobs; ++i) { |
1736 | 12.0k | YUVToRGBThreadData * tdata = &threadData[i]; |
1737 | 12.0k | if (tdata->threadCreated && !avifJoinYUVToRGBThread(tdata)) { |
1738 | 0 | result = AVIF_RESULT_REFORMAT_FAILED; |
1739 | 0 | } |
1740 | 12.0k | if (tdata->result != AVIF_RESULT_OK) { |
1741 | 0 | result = tdata->result; |
1742 | 0 | } |
1743 | 12.0k | } |
1744 | 1.51k | avifFree(threadData); |
1745 | 1.51k | return result; |
1746 | 1.51k | } |
1747 | | |
1748 | | // Limited -> Full |
1749 | | // Plan: subtract limited offset, then multiply by ratio of FULLSIZE/LIMITEDSIZE (rounding), then clamp. |
1750 | | // RATIO = (FULLY - 0) / (MAXLIMITEDY - MINLIMITEDY) |
1751 | | // ----------------------------------------- |
1752 | | // ( ( (v - MINLIMITEDY) | subtract limited offset |
1753 | | // * FULLY | multiply numerator of ratio |
1754 | | // ) + ((MAXLIMITEDY - MINLIMITEDY) / 2) | add 0.5 (half of denominator) to round |
1755 | | // ) / (MAXLIMITEDY - MINLIMITEDY) | divide by denominator of ratio |
1756 | | // AVIF_CLAMP(v, 0, FULLY) | clamp to full range |
1757 | | // ----------------------------------------- |
1758 | | #define LIMITED_TO_FULL(MINLIMITEDY, MAXLIMITEDY, FULLY) \ |
1759 | 103k | v = (((v - MINLIMITEDY) * FULLY) + ((MAXLIMITEDY - MINLIMITEDY) / 2)) / (MAXLIMITEDY - MINLIMITEDY); \ |
1760 | 103k | v = AVIF_CLAMP(v, 0, FULLY) |
1761 | | |
1762 | | // Full -> Limited |
1763 | | // Plan: multiply by ratio of LIMITEDSIZE/FULLSIZE (rounding), then add limited offset, then clamp. |
1764 | | // RATIO = (MAXLIMITEDY - MINLIMITEDY) / (FULLY - 0) |
1765 | | // ----------------------------------------- |
1766 | | // ( ( (v * (MAXLIMITEDY - MINLIMITEDY)) | multiply numerator of ratio |
1767 | | // + (FULLY / 2) | add 0.5 (half of denominator) to round |
1768 | | // ) / FULLY | divide by denominator of ratio |
1769 | | // ) + MINLIMITEDY | add limited offset |
1770 | | // AVIF_CLAMP(v, MINLIMITEDY, MAXLIMITEDY) | clamp to limited range |
1771 | | // ----------------------------------------- |
1772 | | #define FULL_TO_LIMITED(MINLIMITEDY, MAXLIMITEDY, FULLY) \ |
1773 | 0 | v = (((v * (MAXLIMITEDY - MINLIMITEDY)) + (FULLY / 2)) / FULLY) + MINLIMITEDY; \ |
1774 | 0 | v = AVIF_CLAMP(v, MINLIMITEDY, MAXLIMITEDY) |
1775 | | |
1776 | | int avifLimitedToFullY(uint32_t depth, int v) |
1777 | 103k | { |
1778 | 103k | switch (depth) { |
1779 | 88.8k | case 8: |
1780 | 88.8k | LIMITED_TO_FULL(16, 235, 255); |
1781 | 88.8k | break; |
1782 | 15.0k | case 10: |
1783 | 15.0k | LIMITED_TO_FULL(64, 940, 1023); |
1784 | 15.0k | break; |
1785 | 0 | case 12: |
1786 | 0 | LIMITED_TO_FULL(256, 3760, 4095); |
1787 | 0 | break; |
1788 | 103k | } |
1789 | 103k | return v; |
1790 | 103k | } |
1791 | | |
1792 | | int avifLimitedToFullUV(uint32_t depth, int v) |
1793 | 0 | { |
1794 | 0 | switch (depth) { |
1795 | 0 | case 8: |
1796 | 0 | LIMITED_TO_FULL(16, 240, 255); |
1797 | 0 | break; |
1798 | 0 | case 10: |
1799 | 0 | LIMITED_TO_FULL(64, 960, 1023); |
1800 | 0 | break; |
1801 | 0 | case 12: |
1802 | 0 | LIMITED_TO_FULL(256, 3840, 4095); |
1803 | 0 | break; |
1804 | 0 | } |
1805 | 0 | return v; |
1806 | 0 | } |
1807 | | |
1808 | | int avifFullToLimitedY(uint32_t depth, int v) |
1809 | 0 | { |
1810 | 0 | switch (depth) { |
1811 | 0 | case 8: |
1812 | 0 | FULL_TO_LIMITED(16, 235, 255); |
1813 | 0 | break; |
1814 | 0 | case 10: |
1815 | 0 | FULL_TO_LIMITED(64, 940, 1023); |
1816 | 0 | break; |
1817 | 0 | case 12: |
1818 | 0 | FULL_TO_LIMITED(256, 3760, 4095); |
1819 | 0 | break; |
1820 | 0 | } |
1821 | 0 | return v; |
1822 | 0 | } |
1823 | | |
1824 | | int avifFullToLimitedUV(uint32_t depth, int v) |
1825 | 0 | { |
1826 | 0 | switch (depth) { |
1827 | 0 | case 8: |
1828 | 0 | FULL_TO_LIMITED(16, 240, 255); |
1829 | 0 | break; |
1830 | 0 | case 10: |
1831 | 0 | FULL_TO_LIMITED(64, 960, 1023); |
1832 | 0 | break; |
1833 | 0 | case 12: |
1834 | 0 | FULL_TO_LIMITED(256, 3840, 4095); |
1835 | 0 | break; |
1836 | 0 | } |
1837 | 0 | return v; |
1838 | 0 | } |
1839 | | |
1840 | | static inline uint16_t avifFloatToF16(float v) |
1841 | 0 | { |
1842 | 0 | avifF16 f16; |
1843 | 0 | f16.f = v * F16_MULTIPLIER; |
1844 | 0 | return (uint16_t)(f16.u32 >> 13); |
1845 | 0 | } |
1846 | | |
1847 | | static inline float avifF16ToFloat(uint16_t v) |
1848 | 0 | { |
1849 | 0 | avifF16 f16; |
1850 | 0 | f16.u32 = v << 13; |
1851 | 0 | return f16.f / F16_MULTIPLIER; |
1852 | 0 | } |
1853 | | |
1854 | | void avifGetRGBAPixel(const avifRGBImage * src, uint32_t x, uint32_t y, const avifRGBColorSpaceInfo * info, float rgbaPixel[4]) |
1855 | 0 | { |
1856 | 0 | assert(src != NULL); |
1857 | 0 | assert(!src->isFloat || src->depth == 16); |
1858 | 0 | assert(src->format != AVIF_RGB_FORMAT_RGB_565 || src->depth == 8); |
1859 | |
|
1860 | 0 | const uint8_t * const srcPixel = &src->pixels[(size_t)y * src->rowBytes + (size_t)x * info->pixelBytes]; |
1861 | 0 | if (info->channelBytes > 1) { |
1862 | 0 | uint16_t r = *((const uint16_t *)(&srcPixel[info->offsetBytesR])); |
1863 | 0 | uint16_t g = *((const uint16_t *)(&srcPixel[info->offsetBytesG])); |
1864 | 0 | uint16_t b = *((const uint16_t *)(&srcPixel[info->offsetBytesB])); |
1865 | 0 | uint16_t a = avifRGBFormatHasAlpha(src->format) ? *((const uint16_t *)(&srcPixel[info->offsetBytesA])) : (uint16_t)info->maxChannel; |
1866 | 0 | if (src->isFloat) { |
1867 | 0 | rgbaPixel[0] = avifF16ToFloat(r); |
1868 | 0 | rgbaPixel[1] = avifF16ToFloat(g); |
1869 | 0 | rgbaPixel[2] = avifF16ToFloat(b); |
1870 | 0 | rgbaPixel[3] = avifRGBFormatHasAlpha(src->format) ? avifF16ToFloat(a) : 1.0f; |
1871 | 0 | } else { |
1872 | 0 | rgbaPixel[0] = r / info->maxChannelF; |
1873 | 0 | rgbaPixel[1] = g / info->maxChannelF; |
1874 | 0 | rgbaPixel[2] = b / info->maxChannelF; |
1875 | 0 | rgbaPixel[3] = a / info->maxChannelF; |
1876 | 0 | } |
1877 | 0 | } else { |
1878 | 0 | if (src->format == AVIF_RGB_FORMAT_RGB_565) { |
1879 | 0 | uint8_t r, g, b; |
1880 | 0 | avifGetRGB565(&srcPixel[info->offsetBytesR], &r, &g, &b); |
1881 | 0 | rgbaPixel[0] = r / info->maxChannelF; |
1882 | 0 | rgbaPixel[1] = g / info->maxChannelF; |
1883 | 0 | rgbaPixel[2] = b / info->maxChannelF; |
1884 | 0 | rgbaPixel[3] = 1.0f; |
1885 | 0 | } else { |
1886 | 0 | rgbaPixel[0] = srcPixel[info->offsetBytesR] / info->maxChannelF; |
1887 | 0 | rgbaPixel[1] = srcPixel[info->offsetBytesG] / info->maxChannelF; |
1888 | 0 | rgbaPixel[2] = srcPixel[info->offsetBytesB] / info->maxChannelF; |
1889 | 0 | rgbaPixel[3] = avifRGBFormatHasAlpha(src->format) ? (srcPixel[info->offsetBytesA] / info->maxChannelF) : 1.0f; |
1890 | 0 | } |
1891 | 0 | } |
1892 | 0 | } |
1893 | | |
1894 | | void avifSetRGBAPixel(const avifRGBImage * dst, uint32_t x, uint32_t y, const avifRGBColorSpaceInfo * info, const float rgbaPixel[4]) |
1895 | 0 | { |
1896 | 0 | assert(dst != NULL); |
1897 | 0 | assert(!dst->isFloat || dst->depth == 16); |
1898 | 0 | assert(dst->format != AVIF_RGB_FORMAT_RGB_565 || dst->depth == 8); |
1899 | 0 | assert(rgbaPixel[0] >= 0.0f && rgbaPixel[0] <= 1.0f); |
1900 | 0 | assert(rgbaPixel[1] >= 0.0f && rgbaPixel[1] <= 1.0f); |
1901 | 0 | assert(rgbaPixel[2] >= 0.0f && rgbaPixel[2] <= 1.0f); |
1902 | |
|
1903 | 0 | uint8_t * const dstPixel = &dst->pixels[(size_t)y * dst->rowBytes + (size_t)x * info->pixelBytes]; |
1904 | |
|
1905 | 0 | uint8_t * const ptrR = &dstPixel[info->offsetBytesR]; |
1906 | 0 | uint8_t * const ptrG = &dstPixel[info->offsetBytesG]; |
1907 | 0 | uint8_t * const ptrB = &dstPixel[info->offsetBytesB]; |
1908 | 0 | uint8_t * const ptrA = avifRGBFormatHasAlpha(dst->format) ? &dstPixel[info->offsetBytesA] : NULL; |
1909 | 0 | if (dst->depth > 8) { |
1910 | 0 | if (dst->isFloat) { |
1911 | 0 | *((uint16_t *)ptrR) = avifFloatToF16(rgbaPixel[0]); |
1912 | 0 | *((uint16_t *)ptrG) = avifFloatToF16(rgbaPixel[1]); |
1913 | 0 | *((uint16_t *)ptrB) = avifFloatToF16(rgbaPixel[2]); |
1914 | 0 | if (ptrA) { |
1915 | 0 | *((uint16_t *)ptrA) = avifFloatToF16(rgbaPixel[3]); |
1916 | 0 | } |
1917 | 0 | } else { |
1918 | 0 | *((uint16_t *)ptrR) = (uint16_t)(0.5f + (rgbaPixel[0] * info->maxChannelF)); |
1919 | 0 | *((uint16_t *)ptrG) = (uint16_t)(0.5f + (rgbaPixel[1] * info->maxChannelF)); |
1920 | 0 | *((uint16_t *)ptrB) = (uint16_t)(0.5f + (rgbaPixel[2] * info->maxChannelF)); |
1921 | 0 | if (ptrA) { |
1922 | 0 | *((uint16_t *)ptrA) = (uint16_t)(0.5f + (rgbaPixel[3] * info->maxChannelF)); |
1923 | 0 | } |
1924 | 0 | } |
1925 | 0 | } else { |
1926 | 0 | avifStoreRGB8Pixel(dst->format, |
1927 | 0 | (uint8_t)(0.5f + (rgbaPixel[0] * info->maxChannelF)), |
1928 | 0 | (uint8_t)(0.5f + (rgbaPixel[1] * info->maxChannelF)), |
1929 | 0 | (uint8_t)(0.5f + (rgbaPixel[2] * info->maxChannelF)), |
1930 | 0 | ptrR, |
1931 | 0 | ptrG, |
1932 | 0 | ptrB); |
1933 | 0 | if (ptrA) { |
1934 | 0 | *ptrA = (uint8_t)(0.5f + (rgbaPixel[3] * info->maxChannelF)); |
1935 | 0 | } |
1936 | 0 | } |
1937 | 0 | } |